1 /*
2  * Copyright (C) 2015 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #define TRACE_TAG SYSDEPS
18 
19 #include "sysdeps.h"
20 
21 #include <winsock2.h> /* winsock.h *must* be included before windows.h. */
22 #include <windows.h>
23 
24 #include <errno.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 
28 #include <algorithm>
29 #include <memory>
30 #include <mutex>
31 #include <string>
32 #include <unordered_map>
33 #include <vector>
34 
35 #include <cutils/sockets.h>
36 
37 #include <android-base/errors.h>
38 #include <android-base/logging.h>
39 #include <android-base/stringprintf.h>
40 #include <android-base/strings.h>
41 #include <android-base/utf8.h>
42 
43 #include "adb.h"
44 #include "adb_utils.h"
45 
46 extern void fatal(const char *fmt, ...);
47 
48 /* forward declarations */
49 
50 typedef const struct FHClassRec_* FHClass;
51 typedef struct FHRec_* FH;
52 typedef struct EventHookRec_* EventHook;
53 
54 typedef struct FHClassRec_ {
55     void (*_fh_init)(FH);
56     int (*_fh_close)(FH);
57     int (*_fh_lseek)(FH, int, int);
58     int (*_fh_read)(FH, void*, int);
59     int (*_fh_write)(FH, const void*, int);
60 } FHClassRec;
61 
62 static void _fh_file_init(FH);
63 static int _fh_file_close(FH);
64 static int _fh_file_lseek(FH, int, int);
65 static int _fh_file_read(FH, void*, int);
66 static int _fh_file_write(FH, const void*, int);
67 
68 static const FHClassRec _fh_file_class = {
69     _fh_file_init,
70     _fh_file_close,
71     _fh_file_lseek,
72     _fh_file_read,
73     _fh_file_write,
74 };
75 
76 static void _fh_socket_init(FH);
77 static int _fh_socket_close(FH);
78 static int _fh_socket_lseek(FH, int, int);
79 static int _fh_socket_read(FH, void*, int);
80 static int _fh_socket_write(FH, const void*, int);
81 
82 static const FHClassRec _fh_socket_class = {
83     _fh_socket_init,
84     _fh_socket_close,
85     _fh_socket_lseek,
86     _fh_socket_read,
87     _fh_socket_write,
88 };
89 
90 #define assert(cond)                                                                       \
91     do {                                                                                   \
92         if (!(cond)) fatal("assertion failed '%s' on %s:%d\n", #cond, __FILE__, __LINE__); \
93     } while (0)
94 
operator ()(HANDLE h)95 void handle_deleter::operator()(HANDLE h) {
96     // CreateFile() is documented to return INVALID_HANDLE_FILE on error,
97     // implying that NULL is a valid handle, but this is probably impossible.
98     // Other APIs like CreateEvent() are documented to return NULL on error,
99     // implying that INVALID_HANDLE_VALUE is a valid handle, but this is also
100     // probably impossible. Thus, consider both NULL and INVALID_HANDLE_VALUE
101     // as invalid handles. std::unique_ptr won't call a deleter with NULL, so we
102     // only need to check for INVALID_HANDLE_VALUE.
103     if (h != INVALID_HANDLE_VALUE) {
104         if (!CloseHandle(h)) {
105             D("CloseHandle(%p) failed: %s", h,
106               android::base::SystemErrorCodeToString(GetLastError()).c_str());
107         }
108     }
109 }
110 
111 /**************************************************************************/
112 /**************************************************************************/
113 /*****                                                                *****/
114 /*****    common file descriptor handling                             *****/
115 /*****                                                                *****/
116 /**************************************************************************/
117 /**************************************************************************/
118 
119 typedef struct FHRec_
120 {
121     FHClass    clazz;
122     int        used;
123     int        eof;
124     union {
125         HANDLE      handle;
126         SOCKET      socket;
127     } u;
128 
129     char  name[32];
130 } FHRec;
131 
132 #define  fh_handle  u.handle
133 #define  fh_socket  u.socket
134 
135 #define  WIN32_FH_BASE    2048
136 #define  WIN32_MAX_FHS    2048
137 
138 static  std::mutex&  _win32_lock = *new std::mutex();
139 static  FHRec        _win32_fhs[ WIN32_MAX_FHS ];
140 static  int          _win32_fh_next;  // where to start search for free FHRec
141 
142 static FH
_fh_from_int(int fd,const char * func)143 _fh_from_int( int   fd, const char*   func )
144 {
145     FH  f;
146 
147     fd -= WIN32_FH_BASE;
148 
149     if (fd < 0 || fd >= WIN32_MAX_FHS) {
150         D( "_fh_from_int: invalid fd %d passed to %s", fd + WIN32_FH_BASE,
151            func );
152         errno = EBADF;
153         return NULL;
154     }
155 
156     f = &_win32_fhs[fd];
157 
158     if (f->used == 0) {
159         D( "_fh_from_int: invalid fd %d passed to %s", fd + WIN32_FH_BASE,
160            func );
161         errno = EBADF;
162         return NULL;
163     }
164 
165     return f;
166 }
167 
168 
169 static int
_fh_to_int(FH f)170 _fh_to_int( FH  f )
171 {
172     if (f && f->used && f >= _win32_fhs && f < _win32_fhs + WIN32_MAX_FHS)
173         return (int)(f - _win32_fhs) + WIN32_FH_BASE;
174 
175     return -1;
176 }
177 
178 static FH
_fh_alloc(FHClass clazz)179 _fh_alloc( FHClass  clazz )
180 {
181     FH   f = NULL;
182 
183     std::lock_guard<std::mutex> lock(_win32_lock);
184 
185     for (int i = _win32_fh_next; i < WIN32_MAX_FHS; ++i) {
186         if (_win32_fhs[i].clazz == NULL) {
187             f = &_win32_fhs[i];
188             _win32_fh_next = i + 1;
189             f->clazz = clazz;
190             f->used = 1;
191             f->eof = 0;
192             f->name[0] = '\0';
193             clazz->_fh_init(f);
194             return f;
195         }
196     }
197 
198     D("_fh_alloc: no more free file descriptors");
199     errno = EMFILE;  // Too many open files
200     return nullptr;
201 }
202 
203 
204 static int
_fh_close(FH f)205 _fh_close( FH   f )
206 {
207     // Use lock so that closing only happens once and so that _fh_alloc can't
208     // allocate a FH that we're in the middle of closing.
209     std::lock_guard<std::mutex> lock(_win32_lock);
210 
211     int offset = f - _win32_fhs;
212     if (_win32_fh_next > offset) {
213         _win32_fh_next = offset;
214     }
215 
216     if (f->used) {
217         f->clazz->_fh_close( f );
218         f->name[0] = '\0';
219         f->eof     = 0;
220         f->used    = 0;
221         f->clazz   = NULL;
222     }
223     return 0;
224 }
225 
226 // Deleter for unique_fh.
227 class fh_deleter {
228  public:
operator ()(struct FHRec_ * fh)229   void operator()(struct FHRec_* fh) {
230     // We're called from a destructor and destructors should not overwrite
231     // errno because callers may do:
232     //   errno = EBLAH;
233     //   return -1; // calls destructor, which should not overwrite errno
234     const int saved_errno = errno;
235     _fh_close(fh);
236     errno = saved_errno;
237   }
238 };
239 
240 // Like std::unique_ptr, but calls _fh_close() instead of operator delete().
241 typedef std::unique_ptr<struct FHRec_, fh_deleter> unique_fh;
242 
243 /**************************************************************************/
244 /**************************************************************************/
245 /*****                                                                *****/
246 /*****    file-based descriptor handling                              *****/
247 /*****                                                                *****/
248 /**************************************************************************/
249 /**************************************************************************/
250 
_fh_file_init(FH f)251 static void _fh_file_init( FH  f ) {
252     f->fh_handle = INVALID_HANDLE_VALUE;
253 }
254 
_fh_file_close(FH f)255 static int _fh_file_close( FH  f ) {
256     CloseHandle( f->fh_handle );
257     f->fh_handle = INVALID_HANDLE_VALUE;
258     return 0;
259 }
260 
_fh_file_read(FH f,void * buf,int len)261 static int _fh_file_read( FH  f,  void*  buf, int   len ) {
262     DWORD  read_bytes;
263 
264     if ( !ReadFile( f->fh_handle, buf, (DWORD)len, &read_bytes, NULL ) ) {
265         D( "adb_read: could not read %d bytes from %s", len, f->name );
266         errno = EIO;
267         return -1;
268     } else if (read_bytes < (DWORD)len) {
269         f->eof = 1;
270     }
271     return (int)read_bytes;
272 }
273 
_fh_file_write(FH f,const void * buf,int len)274 static int _fh_file_write( FH  f,  const void*  buf, int   len ) {
275     DWORD  wrote_bytes;
276 
277     if ( !WriteFile( f->fh_handle, buf, (DWORD)len, &wrote_bytes, NULL ) ) {
278         D( "adb_file_write: could not write %d bytes from %s", len, f->name );
279         errno = EIO;
280         return -1;
281     } else if (wrote_bytes < (DWORD)len) {
282         f->eof = 1;
283     }
284     return  (int)wrote_bytes;
285 }
286 
_fh_file_lseek(FH f,int pos,int origin)287 static int _fh_file_lseek( FH  f, int  pos, int  origin ) {
288     DWORD  method;
289     DWORD  result;
290 
291     switch (origin)
292     {
293         case SEEK_SET:  method = FILE_BEGIN; break;
294         case SEEK_CUR:  method = FILE_CURRENT; break;
295         case SEEK_END:  method = FILE_END; break;
296         default:
297             errno = EINVAL;
298             return -1;
299     }
300 
301     result = SetFilePointer( f->fh_handle, pos, NULL, method );
302     if (result == INVALID_SET_FILE_POINTER) {
303         errno = EIO;
304         return -1;
305     } else {
306         f->eof = 0;
307     }
308     return (int)result;
309 }
310 
311 
312 /**************************************************************************/
313 /**************************************************************************/
314 /*****                                                                *****/
315 /*****    file-based descriptor handling                              *****/
316 /*****                                                                *****/
317 /**************************************************************************/
318 /**************************************************************************/
319 
adb_open(const char * path,int options)320 int  adb_open(const char*  path, int  options)
321 {
322     FH  f;
323 
324     DWORD  desiredAccess       = 0;
325     DWORD  shareMode           = FILE_SHARE_READ | FILE_SHARE_WRITE;
326 
327     switch (options) {
328         case O_RDONLY:
329             desiredAccess = GENERIC_READ;
330             break;
331         case O_WRONLY:
332             desiredAccess = GENERIC_WRITE;
333             break;
334         case O_RDWR:
335             desiredAccess = GENERIC_READ | GENERIC_WRITE;
336             break;
337         default:
338             D("adb_open: invalid options (0x%0x)", options);
339             errno = EINVAL;
340             return -1;
341     }
342 
343     f = _fh_alloc( &_fh_file_class );
344     if ( !f ) {
345         return -1;
346     }
347 
348     std::wstring path_wide;
349     if (!android::base::UTF8ToWide(path, &path_wide)) {
350         return -1;
351     }
352     f->fh_handle = CreateFileW( path_wide.c_str(), desiredAccess, shareMode,
353                                 NULL, OPEN_EXISTING, 0, NULL );
354 
355     if ( f->fh_handle == INVALID_HANDLE_VALUE ) {
356         const DWORD err = GetLastError();
357         _fh_close(f);
358         D( "adb_open: could not open '%s': ", path );
359         switch (err) {
360             case ERROR_FILE_NOT_FOUND:
361                 D( "file not found" );
362                 errno = ENOENT;
363                 return -1;
364 
365             case ERROR_PATH_NOT_FOUND:
366                 D( "path not found" );
367                 errno = ENOTDIR;
368                 return -1;
369 
370             default:
371                 D("unknown error: %s", android::base::SystemErrorCodeToString(err).c_str());
372                 errno = ENOENT;
373                 return -1;
374         }
375     }
376 
377     snprintf( f->name, sizeof(f->name), "%d(%s)", _fh_to_int(f), path );
378     D( "adb_open: '%s' => fd %d", path, _fh_to_int(f) );
379     return _fh_to_int(f);
380 }
381 
382 /* ignore mode on Win32 */
adb_creat(const char * path,int mode)383 int  adb_creat(const char*  path, int  mode)
384 {
385     FH  f;
386 
387     f = _fh_alloc( &_fh_file_class );
388     if ( !f ) {
389         return -1;
390     }
391 
392     std::wstring path_wide;
393     if (!android::base::UTF8ToWide(path, &path_wide)) {
394         return -1;
395     }
396     f->fh_handle = CreateFileW( path_wide.c_str(), GENERIC_WRITE,
397                                 FILE_SHARE_READ | FILE_SHARE_WRITE,
398                                 NULL, CREATE_ALWAYS, FILE_ATTRIBUTE_NORMAL,
399                                 NULL );
400 
401     if ( f->fh_handle == INVALID_HANDLE_VALUE ) {
402         const DWORD err = GetLastError();
403         _fh_close(f);
404         D( "adb_creat: could not open '%s': ", path );
405         switch (err) {
406             case ERROR_FILE_NOT_FOUND:
407                 D( "file not found" );
408                 errno = ENOENT;
409                 return -1;
410 
411             case ERROR_PATH_NOT_FOUND:
412                 D( "path not found" );
413                 errno = ENOTDIR;
414                 return -1;
415 
416             default:
417                 D("unknown error: %s", android::base::SystemErrorCodeToString(err).c_str());
418                 errno = ENOENT;
419                 return -1;
420         }
421     }
422     snprintf( f->name, sizeof(f->name), "%d(%s)", _fh_to_int(f), path );
423     D( "adb_creat: '%s' => fd %d", path, _fh_to_int(f) );
424     return _fh_to_int(f);
425 }
426 
427 
adb_read(int fd,void * buf,int len)428 int  adb_read(int  fd, void* buf, int len)
429 {
430     FH     f = _fh_from_int(fd, __func__);
431 
432     if (f == NULL) {
433         return -1;
434     }
435 
436     return f->clazz->_fh_read( f, buf, len );
437 }
438 
439 
adb_write(int fd,const void * buf,int len)440 int  adb_write(int  fd, const void*  buf, int  len)
441 {
442     FH     f = _fh_from_int(fd, __func__);
443 
444     if (f == NULL) {
445         return -1;
446     }
447 
448     return f->clazz->_fh_write(f, buf, len);
449 }
450 
451 
adb_lseek(int fd,int pos,int where)452 int  adb_lseek(int  fd, int  pos, int  where)
453 {
454     FH     f = _fh_from_int(fd, __func__);
455 
456     if (!f) {
457         return -1;
458     }
459 
460     return f->clazz->_fh_lseek(f, pos, where);
461 }
462 
463 
adb_close(int fd)464 int  adb_close(int  fd)
465 {
466     FH   f = _fh_from_int(fd, __func__);
467 
468     if (!f) {
469         return -1;
470     }
471 
472     D( "adb_close: %s", f->name);
473     _fh_close(f);
474     return 0;
475 }
476 
477 /**************************************************************************/
478 /**************************************************************************/
479 /*****                                                                *****/
480 /*****    socket-based file descriptors                               *****/
481 /*****                                                                *****/
482 /**************************************************************************/
483 /**************************************************************************/
484 
485 #undef setsockopt
486 
_socket_set_errno(const DWORD err)487 static void _socket_set_errno( const DWORD err ) {
488     // Because the Windows C Runtime (MSVCRT.DLL) strerror() does not support a
489     // lot of POSIX and socket error codes, some of the resulting error codes
490     // are mapped to strings by adb_strerror().
491     switch ( err ) {
492     case 0:              errno = 0; break;
493     // Don't map WSAEINTR since that is only for Winsock 1.1 which we don't use.
494     // case WSAEINTR:    errno = EINTR; break;
495     case WSAEFAULT:      errno = EFAULT; break;
496     case WSAEINVAL:      errno = EINVAL; break;
497     case WSAEMFILE:      errno = EMFILE; break;
498     // Mapping WSAEWOULDBLOCK to EAGAIN is absolutely critical because
499     // non-blocking sockets can cause an error code of WSAEWOULDBLOCK and
500     // callers check specifically for EAGAIN.
501     case WSAEWOULDBLOCK: errno = EAGAIN; break;
502     case WSAENOTSOCK:    errno = ENOTSOCK; break;
503     case WSAENOPROTOOPT: errno = ENOPROTOOPT; break;
504     case WSAEOPNOTSUPP:  errno = EOPNOTSUPP; break;
505     case WSAENETDOWN:    errno = ENETDOWN; break;
506     case WSAENETRESET:   errno = ENETRESET; break;
507     // Map WSAECONNABORTED to EPIPE instead of ECONNABORTED because POSIX seems
508     // to use EPIPE for these situations and there are some callers that look
509     // for EPIPE.
510     case WSAECONNABORTED: errno = EPIPE; break;
511     case WSAECONNRESET:  errno = ECONNRESET; break;
512     case WSAENOBUFS:     errno = ENOBUFS; break;
513     case WSAENOTCONN:    errno = ENOTCONN; break;
514     // Don't map WSAETIMEDOUT because we don't currently use SO_RCVTIMEO or
515     // SO_SNDTIMEO which would cause WSAETIMEDOUT to be returned. Future
516     // considerations: Reportedly send() can return zero on timeout, and POSIX
517     // code may expect EAGAIN instead of ETIMEDOUT on timeout.
518     // case WSAETIMEDOUT: errno = ETIMEDOUT; break;
519     case WSAEHOSTUNREACH: errno = EHOSTUNREACH; break;
520     default:
521         errno = EINVAL;
522         D( "_socket_set_errno: mapping Windows error code %lu to errno %d",
523            err, errno );
524     }
525 }
526 
adb_poll(adb_pollfd * fds,size_t nfds,int timeout)527 extern int adb_poll(adb_pollfd* fds, size_t nfds, int timeout) {
528     // WSAPoll doesn't handle invalid/non-socket handles, so we need to handle them ourselves.
529     int skipped = 0;
530     std::vector<WSAPOLLFD> sockets;
531     std::vector<adb_pollfd*> original;
532     for (size_t i = 0; i < nfds; ++i) {
533         FH fh = _fh_from_int(fds[i].fd, __func__);
534         if (!fh || !fh->used || fh->clazz != &_fh_socket_class) {
535             D("adb_poll received bad FD %d", fds[i].fd);
536             fds[i].revents = POLLNVAL;
537             ++skipped;
538         } else {
539             WSAPOLLFD wsapollfd = {
540                 .fd = fh->u.socket,
541                 .events = static_cast<short>(fds[i].events)
542             };
543             sockets.push_back(wsapollfd);
544             original.push_back(&fds[i]);
545         }
546     }
547 
548     if (sockets.empty()) {
549         return skipped;
550     }
551 
552     int result = WSAPoll(sockets.data(), sockets.size(), timeout);
553     if (result == SOCKET_ERROR) {
554         _socket_set_errno(WSAGetLastError());
555         return -1;
556     }
557 
558     // Map the results back onto the original set.
559     for (size_t i = 0; i < sockets.size(); ++i) {
560         original[i]->revents = sockets[i].revents;
561     }
562 
563     // WSAPoll appears to return the number of unique FDs with avaiable events, instead of how many
564     // of the pollfd elements have a non-zero revents field, which is what it and poll are specified
565     // to do. Ignore its result and calculate the proper return value.
566     result = 0;
567     for (size_t i = 0; i < nfds; ++i) {
568         if (fds[i].revents != 0) {
569             ++result;
570         }
571     }
572     return result;
573 }
574 
_fh_socket_init(FH f)575 static void _fh_socket_init(FH f) {
576     f->fh_socket = INVALID_SOCKET;
577 }
578 
_fh_socket_close(FH f)579 static int _fh_socket_close( FH  f ) {
580     if (f->fh_socket != INVALID_SOCKET) {
581         /* gently tell any peer that we're closing the socket */
582         if (shutdown(f->fh_socket, SD_BOTH) == SOCKET_ERROR) {
583             // If the socket is not connected, this returns an error. We want to
584             // minimize logging spam, so don't log these errors for now.
585 #if 0
586             D("socket shutdown failed: %s",
587               android::base::SystemErrorCodeToString(WSAGetLastError()).c_str());
588 #endif
589         }
590         if (closesocket(f->fh_socket) == SOCKET_ERROR) {
591             // Don't set errno here, since adb_close will ignore it.
592             const DWORD err = WSAGetLastError();
593             D("closesocket failed: %s", android::base::SystemErrorCodeToString(err).c_str());
594         }
595         f->fh_socket = INVALID_SOCKET;
596     }
597     return 0;
598 }
599 
_fh_socket_lseek(FH f,int pos,int origin)600 static int _fh_socket_lseek( FH  f, int pos, int origin ) {
601     errno = EPIPE;
602     return -1;
603 }
604 
_fh_socket_read(FH f,void * buf,int len)605 static int _fh_socket_read(FH f, void* buf, int len) {
606     int  result = recv(f->fh_socket, reinterpret_cast<char*>(buf), len, 0);
607     if (result == SOCKET_ERROR) {
608         const DWORD err = WSAGetLastError();
609         // WSAEWOULDBLOCK is normal with a non-blocking socket, so don't trace
610         // that to reduce spam and confusion.
611         if (err != WSAEWOULDBLOCK) {
612             D("recv fd %d failed: %s", _fh_to_int(f),
613               android::base::SystemErrorCodeToString(err).c_str());
614         }
615         _socket_set_errno(err);
616         result = -1;
617     }
618     return  result;
619 }
620 
_fh_socket_write(FH f,const void * buf,int len)621 static int _fh_socket_write(FH f, const void* buf, int len) {
622     int  result = send(f->fh_socket, reinterpret_cast<const char*>(buf), len, 0);
623     if (result == SOCKET_ERROR) {
624         const DWORD err = WSAGetLastError();
625         // WSAEWOULDBLOCK is normal with a non-blocking socket, so don't trace
626         // that to reduce spam and confusion.
627         if (err != WSAEWOULDBLOCK) {
628             D("send fd %d failed: %s", _fh_to_int(f),
629               android::base::SystemErrorCodeToString(err).c_str());
630         }
631         _socket_set_errno(err);
632         result = -1;
633     } else {
634         // According to https://code.google.com/p/chromium/issues/detail?id=27870
635         // Winsock Layered Service Providers may cause this.
636         CHECK_LE(result, len) << "Tried to write " << len << " bytes to "
637                               << f->name << ", but " << result
638                               << " bytes reportedly written";
639     }
640     return result;
641 }
642 
643 /**************************************************************************/
644 /**************************************************************************/
645 /*****                                                                *****/
646 /*****    replacement for libs/cutils/socket_xxxx.c                   *****/
647 /*****                                                                *****/
648 /**************************************************************************/
649 /**************************************************************************/
650 
651 #include <winsock2.h>
652 
653 static int  _winsock_init;
654 
655 static void
_init_winsock(void)656 _init_winsock( void )
657 {
658     // TODO: Multiple threads calling this may potentially cause multiple calls
659     // to WSAStartup() which offers no real benefit.
660     if (!_winsock_init) {
661         WSADATA  wsaData;
662         int      rc = WSAStartup( MAKEWORD(2,2), &wsaData);
663         if (rc != 0) {
664             fatal("adb: could not initialize Winsock: %s",
665                   android::base::SystemErrorCodeToString(rc).c_str());
666         }
667         _winsock_init = 1;
668 
669         // Note that we do not call atexit() to register WSACleanup to be called
670         // at normal process termination because:
671         // 1) When exit() is called, there are still threads actively using
672         //    Winsock because we don't cleanly shutdown all threads, so it
673         //    doesn't make sense to call WSACleanup() and may cause problems
674         //    with those threads.
675         // 2) A deadlock can occur when exit() holds a C Runtime lock, then it
676         //    calls WSACleanup() which tries to unload a DLL, which tries to
677         //    grab the LoaderLock. This conflicts with the device_poll_thread
678         //    which holds the LoaderLock because AdbWinApi.dll calls
679         //    setupapi.dll which tries to load wintrust.dll which tries to load
680         //    crypt32.dll which calls atexit() which tries to acquire the C
681         //    Runtime lock that the other thread holds.
682     }
683 }
684 
685 // Map a socket type to an explicit socket protocol instead of using the socket
686 // protocol of 0. Explicit socket protocols are used by most apps and we should
687 // do the same to reduce the chance of exercising uncommon code-paths that might
688 // have problems or that might load different Winsock service providers that
689 // have problems.
GetSocketProtocolFromSocketType(int type)690 static int GetSocketProtocolFromSocketType(int type) {
691     switch (type) {
692         case SOCK_STREAM:
693             return IPPROTO_TCP;
694         case SOCK_DGRAM:
695             return IPPROTO_UDP;
696         default:
697             LOG(FATAL) << "Unknown socket type: " << type;
698             return 0;
699     }
700 }
701 
network_loopback_client(int port,int type,std::string * error)702 int network_loopback_client(int port, int type, std::string* error) {
703     struct sockaddr_in addr;
704     SOCKET s;
705 
706     unique_fh f(_fh_alloc(&_fh_socket_class));
707     if (!f) {
708         *error = strerror(errno);
709         return -1;
710     }
711 
712     if (!_winsock_init) _init_winsock();
713 
714     memset(&addr, 0, sizeof(addr));
715     addr.sin_family = AF_INET;
716     addr.sin_port = htons(port);
717     addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
718 
719     s = socket(AF_INET, type, GetSocketProtocolFromSocketType(type));
720     if (s == INVALID_SOCKET) {
721         const DWORD err = WSAGetLastError();
722         *error = android::base::StringPrintf("cannot create socket: %s",
723                                              android::base::SystemErrorCodeToString(err).c_str());
724         D("%s", error->c_str());
725         _socket_set_errno(err);
726         return -1;
727     }
728     f->fh_socket = s;
729 
730     if (connect(s, (struct sockaddr*)&addr, sizeof(addr)) == SOCKET_ERROR) {
731         // Save err just in case inet_ntoa() or ntohs() changes the last error.
732         const DWORD err = WSAGetLastError();
733         *error = android::base::StringPrintf("cannot connect to %s:%u: %s",
734                                              inet_ntoa(addr.sin_addr), ntohs(addr.sin_port),
735                                              android::base::SystemErrorCodeToString(err).c_str());
736         D("could not connect to %s:%d: %s", type != SOCK_STREAM ? "udp" : "tcp", port,
737           error->c_str());
738         _socket_set_errno(err);
739         return -1;
740     }
741 
742     const int fd = _fh_to_int(f.get());
743     snprintf(f->name, sizeof(f->name), "%d(lo-client:%s%d)", fd, type != SOCK_STREAM ? "udp:" : "",
744              port);
745     D("port %d type %s => fd %d", port, type != SOCK_STREAM ? "udp" : "tcp", fd);
746     f.release();
747     return fd;
748 }
749 
750 #define LISTEN_BACKLOG 4
751 
752 // interface_address is INADDR_LOOPBACK or INADDR_ANY.
_network_server(int port,int type,u_long interface_address,std::string * error)753 static int _network_server(int port, int type, u_long interface_address, std::string* error) {
754     struct sockaddr_in addr;
755     SOCKET s;
756     int n;
757 
758     unique_fh f(_fh_alloc(&_fh_socket_class));
759     if (!f) {
760         *error = strerror(errno);
761         return -1;
762     }
763 
764     if (!_winsock_init) _init_winsock();
765 
766     memset(&addr, 0, sizeof(addr));
767     addr.sin_family = AF_INET;
768     addr.sin_port = htons(port);
769     addr.sin_addr.s_addr = htonl(interface_address);
770 
771     // TODO: Consider using dual-stack socket that can simultaneously listen on
772     // IPv4 and IPv6.
773     s = socket(AF_INET, type, GetSocketProtocolFromSocketType(type));
774     if (s == INVALID_SOCKET) {
775         const DWORD err = WSAGetLastError();
776         *error = android::base::StringPrintf("cannot create socket: %s",
777                                              android::base::SystemErrorCodeToString(err).c_str());
778         D("%s", error->c_str());
779         _socket_set_errno(err);
780         return -1;
781     }
782 
783     f->fh_socket = s;
784 
785     // Note: SO_REUSEADDR on Windows allows multiple processes to bind to the
786     // same port, so instead use SO_EXCLUSIVEADDRUSE.
787     n = 1;
788     if (setsockopt(s, SOL_SOCKET, SO_EXCLUSIVEADDRUSE, (const char*)&n, sizeof(n)) == SOCKET_ERROR) {
789         const DWORD err = WSAGetLastError();
790         *error = android::base::StringPrintf("cannot set socket option SO_EXCLUSIVEADDRUSE: %s",
791                                              android::base::SystemErrorCodeToString(err).c_str());
792         D("%s", error->c_str());
793         _socket_set_errno(err);
794         return -1;
795     }
796 
797     if (bind(s, (struct sockaddr*)&addr, sizeof(addr)) == SOCKET_ERROR) {
798         // Save err just in case inet_ntoa() or ntohs() changes the last error.
799         const DWORD err = WSAGetLastError();
800         *error = android::base::StringPrintf("cannot bind to %s:%u: %s", inet_ntoa(addr.sin_addr),
801                                              ntohs(addr.sin_port),
802                                              android::base::SystemErrorCodeToString(err).c_str());
803         D("could not bind to %s:%d: %s", type != SOCK_STREAM ? "udp" : "tcp", port, error->c_str());
804         _socket_set_errno(err);
805         return -1;
806     }
807     if (type == SOCK_STREAM) {
808         if (listen(s, LISTEN_BACKLOG) == SOCKET_ERROR) {
809             const DWORD err = WSAGetLastError();
810             *error = android::base::StringPrintf(
811                 "cannot listen on socket: %s", android::base::SystemErrorCodeToString(err).c_str());
812             D("could not listen on %s:%d: %s", type != SOCK_STREAM ? "udp" : "tcp", port,
813               error->c_str());
814             _socket_set_errno(err);
815             return -1;
816         }
817     }
818     const int fd = _fh_to_int(f.get());
819     snprintf(f->name, sizeof(f->name), "%d(%s-server:%s%d)", fd,
820              interface_address == INADDR_LOOPBACK ? "lo" : "any", type != SOCK_STREAM ? "udp:" : "",
821              port);
822     D("port %d type %s => fd %d", port, type != SOCK_STREAM ? "udp" : "tcp", fd);
823     f.release();
824     return fd;
825 }
826 
network_loopback_server(int port,int type,std::string * error)827 int network_loopback_server(int port, int type, std::string* error) {
828     return _network_server(port, type, INADDR_LOOPBACK, error);
829 }
830 
network_inaddr_any_server(int port,int type,std::string * error)831 int network_inaddr_any_server(int port, int type, std::string* error) {
832     return _network_server(port, type, INADDR_ANY, error);
833 }
834 
network_connect(const std::string & host,int port,int type,int timeout,std::string * error)835 int network_connect(const std::string& host, int port, int type, int timeout, std::string* error) {
836     unique_fh f(_fh_alloc(&_fh_socket_class));
837     if (!f) {
838         *error = strerror(errno);
839         return -1;
840     }
841 
842     if (!_winsock_init) _init_winsock();
843 
844     struct addrinfo hints;
845     memset(&hints, 0, sizeof(hints));
846     hints.ai_family = AF_UNSPEC;
847     hints.ai_socktype = type;
848     hints.ai_protocol = GetSocketProtocolFromSocketType(type);
849 
850     char port_str[16];
851     snprintf(port_str, sizeof(port_str), "%d", port);
852 
853     struct addrinfo* addrinfo_ptr = nullptr;
854 
855 #if (NTDDI_VERSION >= NTDDI_WINXPSP2) || (_WIN32_WINNT >= _WIN32_WINNT_WS03)
856 // TODO: When the Android SDK tools increases the Windows system
857 // requirements >= WinXP SP2, switch to android::base::UTF8ToWide() + GetAddrInfoW().
858 #else
859 // Otherwise, keep using getaddrinfo(), or do runtime API detection
860 // with GetProcAddress("GetAddrInfoW").
861 #endif
862     if (getaddrinfo(host.c_str(), port_str, &hints, &addrinfo_ptr) != 0) {
863         const DWORD err = WSAGetLastError();
864         *error = android::base::StringPrintf("cannot resolve host '%s' and port %s: %s",
865                                              host.c_str(), port_str,
866                                              android::base::SystemErrorCodeToString(err).c_str());
867 
868         D("%s", error->c_str());
869         _socket_set_errno(err);
870         return -1;
871     }
872     std::unique_ptr<struct addrinfo, decltype(&freeaddrinfo)> addrinfo(addrinfo_ptr, freeaddrinfo);
873     addrinfo_ptr = nullptr;
874 
875     // TODO: Try all the addresses if there's more than one? This just uses
876     // the first. Or, could call WSAConnectByName() (Windows Vista and newer)
877     // which tries all addresses, takes a timeout and more.
878     SOCKET s = socket(addrinfo->ai_family, addrinfo->ai_socktype, addrinfo->ai_protocol);
879     if (s == INVALID_SOCKET) {
880         const DWORD err = WSAGetLastError();
881         *error = android::base::StringPrintf("cannot create socket: %s",
882                                              android::base::SystemErrorCodeToString(err).c_str());
883         D("%s", error->c_str());
884         _socket_set_errno(err);
885         return -1;
886     }
887     f->fh_socket = s;
888 
889     // TODO: Implement timeouts for Windows. Seems like the default in theory
890     // (according to http://serverfault.com/a/671453) and in practice is 21 sec.
891     if (connect(s, addrinfo->ai_addr, addrinfo->ai_addrlen) == SOCKET_ERROR) {
892         // TODO: Use WSAAddressToString or inet_ntop on address.
893         const DWORD err = WSAGetLastError();
894         *error = android::base::StringPrintf("cannot connect to %s:%s: %s", host.c_str(), port_str,
895                                              android::base::SystemErrorCodeToString(err).c_str());
896         D("could not connect to %s:%s:%s: %s", type != SOCK_STREAM ? "udp" : "tcp", host.c_str(),
897           port_str, error->c_str());
898         _socket_set_errno(err);
899         return -1;
900     }
901 
902     const int fd = _fh_to_int(f.get());
903     snprintf(f->name, sizeof(f->name), "%d(net-client:%s%d)", fd, type != SOCK_STREAM ? "udp:" : "",
904              port);
905     D("host '%s' port %d type %s => fd %d", host.c_str(), port, type != SOCK_STREAM ? "udp" : "tcp",
906       fd);
907     f.release();
908     return fd;
909 }
910 
adb_register_socket(SOCKET s)911 int  adb_register_socket(SOCKET s) {
912     FH f = _fh_alloc( &_fh_socket_class );
913     f->fh_socket = s;
914     return _fh_to_int(f);
915 }
916 
917 #undef accept
adb_socket_accept(int serverfd,struct sockaddr * addr,socklen_t * addrlen)918 int  adb_socket_accept(int  serverfd, struct sockaddr*  addr, socklen_t  *addrlen)
919 {
920     FH   serverfh = _fh_from_int(serverfd, __func__);
921 
922     if ( !serverfh || serverfh->clazz != &_fh_socket_class ) {
923         D("adb_socket_accept: invalid fd %d", serverfd);
924         errno = EBADF;
925         return -1;
926     }
927 
928     unique_fh fh(_fh_alloc( &_fh_socket_class ));
929     if (!fh) {
930         PLOG(ERROR) << "adb_socket_accept: failed to allocate accepted socket "
931                        "descriptor";
932         return -1;
933     }
934 
935     fh->fh_socket = accept( serverfh->fh_socket, addr, addrlen );
936     if (fh->fh_socket == INVALID_SOCKET) {
937         const DWORD err = WSAGetLastError();
938         LOG(ERROR) << "adb_socket_accept: accept on fd " << serverfd <<
939                       " failed: " + android::base::SystemErrorCodeToString(err);
940         _socket_set_errno( err );
941         return -1;
942     }
943 
944     const int fd = _fh_to_int(fh.get());
945     snprintf( fh->name, sizeof(fh->name), "%d(accept:%s)", fd, serverfh->name );
946     D( "adb_socket_accept on fd %d returns fd %d", serverfd, fd );
947     fh.release();
948     return  fd;
949 }
950 
951 
adb_setsockopt(int fd,int level,int optname,const void * optval,socklen_t optlen)952 int  adb_setsockopt( int  fd, int  level, int  optname, const void*  optval, socklen_t  optlen )
953 {
954     FH   fh = _fh_from_int(fd, __func__);
955 
956     if ( !fh || fh->clazz != &_fh_socket_class ) {
957         D("adb_setsockopt: invalid fd %d", fd);
958         errno = EBADF;
959         return -1;
960     }
961 
962     // TODO: Once we can assume Windows Vista or later, if the caller is trying
963     // to set SOL_SOCKET, SO_SNDBUF/SO_RCVBUF, ignore it since the OS has
964     // auto-tuning.
965 
966     int result = setsockopt( fh->fh_socket, level, optname,
967                              reinterpret_cast<const char*>(optval), optlen );
968     if ( result == SOCKET_ERROR ) {
969         const DWORD err = WSAGetLastError();
970         D("adb_setsockopt: setsockopt on fd %d level %d optname %d failed: %s\n",
971           fd, level, optname, android::base::SystemErrorCodeToString(err).c_str());
972         _socket_set_errno( err );
973         result = -1;
974     }
975     return result;
976 }
977 
adb_getsockname(int fd,struct sockaddr * sockaddr,socklen_t * optlen)978 int adb_getsockname(int fd, struct sockaddr* sockaddr, socklen_t* optlen) {
979     FH fh = _fh_from_int(fd, __func__);
980 
981     if (!fh || fh->clazz != &_fh_socket_class) {
982         D("adb_getsockname: invalid fd %d", fd);
983         errno = EBADF;
984         return -1;
985     }
986 
987     int result = (getsockname)(fh->fh_socket, sockaddr, optlen);
988     if (result == SOCKET_ERROR) {
989         const DWORD err = WSAGetLastError();
990         D("adb_getsockname: setsockopt on fd %d failed: %s\n", fd,
991           android::base::SystemErrorCodeToString(err).c_str());
992         _socket_set_errno(err);
993         result = -1;
994     }
995     return result;
996 }
997 
adb_socket_get_local_port(int fd)998 int adb_socket_get_local_port(int fd) {
999     sockaddr_storage addr_storage;
1000     socklen_t addr_len = sizeof(addr_storage);
1001 
1002     if (adb_getsockname(fd, reinterpret_cast<sockaddr*>(&addr_storage), &addr_len) < 0) {
1003         D("adb_socket_get_local_port: adb_getsockname failed: %s", strerror(errno));
1004         return -1;
1005     }
1006 
1007     if (!(addr_storage.ss_family == AF_INET || addr_storage.ss_family == AF_INET6)) {
1008         D("adb_socket_get_local_port: unknown address family received: %d", addr_storage.ss_family);
1009         errno = ECONNABORTED;
1010         return -1;
1011     }
1012 
1013     return ntohs(reinterpret_cast<sockaddr_in*>(&addr_storage)->sin_port);
1014 }
1015 
adb_shutdown(int fd)1016 int  adb_shutdown(int  fd)
1017 {
1018     FH   f = _fh_from_int(fd, __func__);
1019 
1020     if (!f || f->clazz != &_fh_socket_class) {
1021         D("adb_shutdown: invalid fd %d", fd);
1022         errno = EBADF;
1023         return -1;
1024     }
1025 
1026     D( "adb_shutdown: %s", f->name);
1027     if (shutdown(f->fh_socket, SD_BOTH) == SOCKET_ERROR) {
1028         const DWORD err = WSAGetLastError();
1029         D("socket shutdown fd %d failed: %s", fd,
1030           android::base::SystemErrorCodeToString(err).c_str());
1031         _socket_set_errno(err);
1032         return -1;
1033     }
1034     return 0;
1035 }
1036 
1037 // Emulate socketpair(2) by binding and connecting to a socket.
adb_socketpair(int sv[2])1038 int adb_socketpair(int sv[2]) {
1039     int server = -1;
1040     int client = -1;
1041     int accepted = -1;
1042     int local_port = -1;
1043     std::string error;
1044 
1045     struct sockaddr_storage peer_addr = {};
1046     struct sockaddr_storage client_addr = {};
1047     socklen_t peer_socklen = sizeof(peer_addr);
1048     socklen_t client_socklen = sizeof(client_addr);
1049 
1050     server = network_loopback_server(0, SOCK_STREAM, &error);
1051     if (server < 0) {
1052         D("adb_socketpair: failed to create server: %s", error.c_str());
1053         goto fail;
1054     }
1055 
1056     local_port = adb_socket_get_local_port(server);
1057     if (local_port < 0) {
1058         D("adb_socketpair: failed to get server port number: %s", error.c_str());
1059         goto fail;
1060     }
1061     D("adb_socketpair: bound on port %d", local_port);
1062 
1063     client = network_loopback_client(local_port, SOCK_STREAM, &error);
1064     if (client < 0) {
1065         D("adb_socketpair: failed to connect client: %s", error.c_str());
1066         goto fail;
1067     }
1068 
1069     // Make sure that the peer that connected to us and the client are the same.
1070     accepted = adb_socket_accept(server, reinterpret_cast<sockaddr*>(&peer_addr), &peer_socklen);
1071     if (accepted < 0) {
1072         D("adb_socketpair: failed to accept: %s", strerror(errno));
1073         goto fail;
1074     }
1075 
1076     if (adb_getsockname(client, reinterpret_cast<sockaddr*>(&client_addr), &client_socklen) != 0) {
1077         D("adb_socketpair: failed to getpeername: %s", strerror(errno));
1078         goto fail;
1079     }
1080 
1081     if (peer_socklen != client_socklen) {
1082         D("adb_socketpair: client and peer sockaddrs have different lengths");
1083         errno = EIO;
1084         goto fail;
1085     }
1086 
1087     if (memcmp(&peer_addr, &client_addr, peer_socklen) != 0) {
1088         D("adb_socketpair: client and peer sockaddrs don't match");
1089         errno = EIO;
1090         goto fail;
1091     }
1092 
1093     adb_close(server);
1094 
1095     sv[0] = client;
1096     sv[1] = accepted;
1097     return 0;
1098 
1099 fail:
1100     if (server >= 0) {
1101         adb_close(server);
1102     }
1103     if (client >= 0) {
1104         adb_close(client);
1105     }
1106     if (accepted >= 0) {
1107         adb_close(accepted);
1108     }
1109     return -1;
1110 }
1111 
set_file_block_mode(int fd,bool block)1112 bool set_file_block_mode(int fd, bool block) {
1113     FH fh = _fh_from_int(fd, __func__);
1114 
1115     if (!fh || !fh->used) {
1116         errno = EBADF;
1117         D("Setting nonblocking on bad file descriptor %d", fd);
1118         return false;
1119     }
1120 
1121     if (fh->clazz == &_fh_socket_class) {
1122         u_long x = !block;
1123         if (ioctlsocket(fh->u.socket, FIONBIO, &x) != 0) {
1124             int error = WSAGetLastError();
1125             _socket_set_errno(error);
1126             D("Setting %d nonblocking failed (%d)", fd, error);
1127             return false;
1128         }
1129         return true;
1130     } else {
1131         errno = ENOTSOCK;
1132         D("Setting nonblocking on non-socket %d", fd);
1133         return false;
1134     }
1135 }
1136 
set_tcp_keepalive(int fd,int interval_sec)1137 bool set_tcp_keepalive(int fd, int interval_sec) {
1138     FH fh = _fh_from_int(fd, __func__);
1139 
1140     if (!fh || fh->clazz != &_fh_socket_class) {
1141         D("set_tcp_keepalive(%d) failed: invalid fd", fd);
1142         errno = EBADF;
1143         return false;
1144     }
1145 
1146     tcp_keepalive keepalive;
1147     keepalive.onoff = (interval_sec > 0);
1148     keepalive.keepalivetime = interval_sec * 1000;
1149     keepalive.keepaliveinterval = interval_sec * 1000;
1150 
1151     DWORD bytes_returned = 0;
1152     if (WSAIoctl(fh->fh_socket, SIO_KEEPALIVE_VALS, &keepalive, sizeof(keepalive), nullptr, 0,
1153                  &bytes_returned, nullptr, nullptr) != 0) {
1154         const DWORD err = WSAGetLastError();
1155         D("set_tcp_keepalive(%d) failed: %s", fd,
1156           android::base::SystemErrorCodeToString(err).c_str());
1157         _socket_set_errno(err);
1158         return false;
1159     }
1160 
1161     return true;
1162 }
1163 
1164 /**************************************************************************/
1165 /**************************************************************************/
1166 /*****                                                                *****/
1167 /*****      Console Window Terminal Emulation                         *****/
1168 /*****                                                                *****/
1169 /**************************************************************************/
1170 /**************************************************************************/
1171 
1172 // This reads input from a Win32 console window and translates it into Unix
1173 // terminal-style sequences. This emulates mostly Gnome Terminal (in Normal
1174 // mode, not Application mode), which itself emulates xterm. Gnome Terminal
1175 // is emulated instead of xterm because it is probably more popular than xterm:
1176 // Ubuntu's default Ctrl-Alt-T shortcut opens Gnome Terminal, Gnome Terminal
1177 // supports modern fonts, etc. It seems best to emulate the terminal that most
1178 // Android developers use because they'll fix apps (the shell, etc.) to keep
1179 // working with that terminal's emulation.
1180 //
1181 // The point of this emulation is not to be perfect or to solve all issues with
1182 // console windows on Windows, but to be better than the original code which
1183 // just called read() (which called ReadFile(), which called ReadConsoleA())
1184 // which did not support Ctrl-C, tab completion, shell input line editing
1185 // keys, server echo, and more.
1186 //
1187 // This implementation reconfigures the console with SetConsoleMode(), then
1188 // calls ReadConsoleInput() to get raw input which it remaps to Unix
1189 // terminal-style sequences which is returned via unix_read() which is used
1190 // by the 'adb shell' command.
1191 //
1192 // Code organization:
1193 //
1194 // * _get_console_handle() and unix_isatty() provide console information.
1195 // * stdin_raw_init() and stdin_raw_restore() reconfigure the console.
1196 // * unix_read() detects console windows (as opposed to pipes, files, etc.).
1197 // * _console_read() is the main code of the emulation.
1198 
1199 // Returns a console HANDLE if |fd| is a console, otherwise returns nullptr.
1200 // If a valid HANDLE is returned and |mode| is not null, |mode| is also filled
1201 // with the console mode. Requires GENERIC_READ access to the underlying HANDLE.
_get_console_handle(int fd,DWORD * mode=nullptr)1202 static HANDLE _get_console_handle(int fd, DWORD* mode=nullptr) {
1203     // First check isatty(); this is very fast and eliminates most non-console
1204     // FDs, but returns 1 for both consoles and character devices like NUL.
1205 #pragma push_macro("isatty")
1206 #undef isatty
1207     if (!isatty(fd)) {
1208         return nullptr;
1209     }
1210 #pragma pop_macro("isatty")
1211 
1212     // To differentiate between character devices and consoles we need to get
1213     // the underlying HANDLE and use GetConsoleMode(), which is what requires
1214     // GENERIC_READ permissions.
1215     const intptr_t intptr_handle = _get_osfhandle(fd);
1216     if (intptr_handle == -1) {
1217         return nullptr;
1218     }
1219     const HANDLE handle = reinterpret_cast<const HANDLE>(intptr_handle);
1220     DWORD temp_mode = 0;
1221     if (!GetConsoleMode(handle, mode ? mode : &temp_mode)) {
1222         return nullptr;
1223     }
1224 
1225     return handle;
1226 }
1227 
1228 // Returns a console handle if |stream| is a console, otherwise returns nullptr.
_get_console_handle(FILE * const stream)1229 static HANDLE _get_console_handle(FILE* const stream) {
1230     // Save and restore errno to make it easier for callers to prevent from overwriting errno.
1231     android::base::ErrnoRestorer er;
1232     const int fd = fileno(stream);
1233     if (fd < 0) {
1234         return nullptr;
1235     }
1236     return _get_console_handle(fd);
1237 }
1238 
unix_isatty(int fd)1239 int unix_isatty(int fd) {
1240     return _get_console_handle(fd) ? 1 : 0;
1241 }
1242 
1243 // Get the next KEY_EVENT_RECORD that should be processed.
_get_key_event_record(const HANDLE console,INPUT_RECORD * const input_record)1244 static bool _get_key_event_record(const HANDLE console, INPUT_RECORD* const input_record) {
1245     for (;;) {
1246         DWORD read_count = 0;
1247         memset(input_record, 0, sizeof(*input_record));
1248         if (!ReadConsoleInputA(console, input_record, 1, &read_count)) {
1249             D("_get_key_event_record: ReadConsoleInputA() failed: %s\n",
1250               android::base::SystemErrorCodeToString(GetLastError()).c_str());
1251             errno = EIO;
1252             return false;
1253         }
1254 
1255         if (read_count == 0) {   // should be impossible
1256             fatal("ReadConsoleInputA returned 0");
1257         }
1258 
1259         if (read_count != 1) {   // should be impossible
1260             fatal("ReadConsoleInputA did not return one input record");
1261         }
1262 
1263         // If the console window is resized, emulate SIGWINCH by breaking out
1264         // of read() with errno == EINTR. Note that there is no event on
1265         // vertical resize because we don't give the console our own custom
1266         // screen buffer (with CreateConsoleScreenBuffer() +
1267         // SetConsoleActiveScreenBuffer()). Instead, we use the default which
1268         // supports scrollback, but doesn't seem to raise an event for vertical
1269         // window resize.
1270         if (input_record->EventType == WINDOW_BUFFER_SIZE_EVENT) {
1271             errno = EINTR;
1272             return false;
1273         }
1274 
1275         if ((input_record->EventType == KEY_EVENT) &&
1276             (input_record->Event.KeyEvent.bKeyDown)) {
1277             if (input_record->Event.KeyEvent.wRepeatCount == 0) {
1278                 fatal("ReadConsoleInputA returned a key event with zero repeat"
1279                       " count");
1280             }
1281 
1282             // Got an interesting INPUT_RECORD, so return
1283             return true;
1284         }
1285     }
1286 }
1287 
_is_shift_pressed(const DWORD control_key_state)1288 static __inline__ bool _is_shift_pressed(const DWORD control_key_state) {
1289     return (control_key_state & SHIFT_PRESSED) != 0;
1290 }
1291 
_is_ctrl_pressed(const DWORD control_key_state)1292 static __inline__ bool _is_ctrl_pressed(const DWORD control_key_state) {
1293     return (control_key_state & (LEFT_CTRL_PRESSED | RIGHT_CTRL_PRESSED)) != 0;
1294 }
1295 
_is_alt_pressed(const DWORD control_key_state)1296 static __inline__ bool _is_alt_pressed(const DWORD control_key_state) {
1297     return (control_key_state & (LEFT_ALT_PRESSED | RIGHT_ALT_PRESSED)) != 0;
1298 }
1299 
_is_numlock_on(const DWORD control_key_state)1300 static __inline__ bool _is_numlock_on(const DWORD control_key_state) {
1301     return (control_key_state & NUMLOCK_ON) != 0;
1302 }
1303 
_is_capslock_on(const DWORD control_key_state)1304 static __inline__ bool _is_capslock_on(const DWORD control_key_state) {
1305     return (control_key_state & CAPSLOCK_ON) != 0;
1306 }
1307 
_is_enhanced_key(const DWORD control_key_state)1308 static __inline__ bool _is_enhanced_key(const DWORD control_key_state) {
1309     return (control_key_state & ENHANCED_KEY) != 0;
1310 }
1311 
1312 // Constants from MSDN for ToAscii().
1313 static const BYTE TOASCII_KEY_OFF = 0x00;
1314 static const BYTE TOASCII_KEY_DOWN = 0x80;
1315 static const BYTE TOASCII_KEY_TOGGLED_ON = 0x01;   // for CapsLock
1316 
1317 // Given a key event, ignore a modifier key and return the character that was
1318 // entered without the modifier. Writes to *ch and returns the number of bytes
1319 // written.
_get_char_ignoring_modifier(char * const ch,const KEY_EVENT_RECORD * const key_event,const DWORD control_key_state,const WORD modifier)1320 static size_t _get_char_ignoring_modifier(char* const ch,
1321     const KEY_EVENT_RECORD* const key_event, const DWORD control_key_state,
1322     const WORD modifier) {
1323     // If there is no character from Windows, try ignoring the specified
1324     // modifier and look for a character. Note that if AltGr is being used,
1325     // there will be a character from Windows.
1326     if (key_event->uChar.AsciiChar == '\0') {
1327         // Note that we read the control key state from the passed in argument
1328         // instead of from key_event since the argument has been normalized.
1329         if (((modifier == VK_SHIFT)   &&
1330             _is_shift_pressed(control_key_state)) ||
1331             ((modifier == VK_CONTROL) &&
1332             _is_ctrl_pressed(control_key_state)) ||
1333             ((modifier == VK_MENU)    && _is_alt_pressed(control_key_state))) {
1334 
1335             BYTE key_state[256]   = {0};
1336             key_state[VK_SHIFT]   = _is_shift_pressed(control_key_state) ?
1337                 TOASCII_KEY_DOWN : TOASCII_KEY_OFF;
1338             key_state[VK_CONTROL] = _is_ctrl_pressed(control_key_state)  ?
1339                 TOASCII_KEY_DOWN : TOASCII_KEY_OFF;
1340             key_state[VK_MENU]    = _is_alt_pressed(control_key_state)   ?
1341                 TOASCII_KEY_DOWN : TOASCII_KEY_OFF;
1342             key_state[VK_CAPITAL] = _is_capslock_on(control_key_state)   ?
1343                 TOASCII_KEY_TOGGLED_ON : TOASCII_KEY_OFF;
1344 
1345             // cause this modifier to be ignored
1346             key_state[modifier]   = TOASCII_KEY_OFF;
1347 
1348             WORD translated = 0;
1349             if (ToAscii(key_event->wVirtualKeyCode,
1350                 key_event->wVirtualScanCode, key_state, &translated, 0) == 1) {
1351                 // Ignoring the modifier, we found a character.
1352                 *ch = (CHAR)translated;
1353                 return 1;
1354             }
1355         }
1356     }
1357 
1358     // Just use whatever Windows told us originally.
1359     *ch = key_event->uChar.AsciiChar;
1360 
1361     // If the character from Windows is NULL, return a size of zero.
1362     return (*ch == '\0') ? 0 : 1;
1363 }
1364 
1365 // If a Ctrl key is pressed, lookup the character, ignoring the Ctrl key,
1366 // but taking into account the shift key. This is because for a sequence like
1367 // Ctrl-Alt-0, we want to find the character '0' and for Ctrl-Alt-Shift-0,
1368 // we want to find the character ')'.
1369 //
1370 // Note that Windows doesn't seem to pass bKeyDown for Ctrl-Shift-NoAlt-0
1371 // because it is the default key-sequence to switch the input language.
1372 // This is configurable in the Region and Language control panel.
_get_non_control_char(char * const ch,const KEY_EVENT_RECORD * const key_event,const DWORD control_key_state)1373 static __inline__ size_t _get_non_control_char(char* const ch,
1374     const KEY_EVENT_RECORD* const key_event, const DWORD control_key_state) {
1375     return _get_char_ignoring_modifier(ch, key_event, control_key_state,
1376         VK_CONTROL);
1377 }
1378 
1379 // Get without Alt.
_get_non_alt_char(char * const ch,const KEY_EVENT_RECORD * const key_event,const DWORD control_key_state)1380 static __inline__ size_t _get_non_alt_char(char* const ch,
1381     const KEY_EVENT_RECORD* const key_event, const DWORD control_key_state) {
1382     return _get_char_ignoring_modifier(ch, key_event, control_key_state,
1383         VK_MENU);
1384 }
1385 
1386 // Ignore the control key, find the character from Windows, and apply any
1387 // Control key mappings (for example, Ctrl-2 is a NULL character). Writes to
1388 // *pch and returns number of bytes written.
_get_control_character(char * const pch,const KEY_EVENT_RECORD * const key_event,const DWORD control_key_state)1389 static size_t _get_control_character(char* const pch,
1390     const KEY_EVENT_RECORD* const key_event, const DWORD control_key_state) {
1391     const size_t len = _get_non_control_char(pch, key_event,
1392         control_key_state);
1393 
1394     if ((len == 1) && _is_ctrl_pressed(control_key_state)) {
1395         char ch = *pch;
1396         switch (ch) {
1397         case '2':
1398         case '@':
1399         case '`':
1400             ch = '\0';
1401             break;
1402         case '3':
1403         case '[':
1404         case '{':
1405             ch = '\x1b';
1406             break;
1407         case '4':
1408         case '\\':
1409         case '|':
1410             ch = '\x1c';
1411             break;
1412         case '5':
1413         case ']':
1414         case '}':
1415             ch = '\x1d';
1416             break;
1417         case '6':
1418         case '^':
1419         case '~':
1420             ch = '\x1e';
1421             break;
1422         case '7':
1423         case '-':
1424         case '_':
1425             ch = '\x1f';
1426             break;
1427         case '8':
1428             ch = '\x7f';
1429             break;
1430         case '/':
1431             if (!_is_alt_pressed(control_key_state)) {
1432                 ch = '\x1f';
1433             }
1434             break;
1435         case '?':
1436             if (!_is_alt_pressed(control_key_state)) {
1437                 ch = '\x7f';
1438             }
1439             break;
1440         }
1441         *pch = ch;
1442     }
1443 
1444     return len;
1445 }
1446 
_normalize_altgr_control_key_state(const KEY_EVENT_RECORD * const key_event)1447 static DWORD _normalize_altgr_control_key_state(
1448     const KEY_EVENT_RECORD* const key_event) {
1449     DWORD control_key_state = key_event->dwControlKeyState;
1450 
1451     // If we're in an AltGr situation where the AltGr key is down (depending on
1452     // the keyboard layout, that might be the physical right alt key which
1453     // produces a control_key_state where Right-Alt and Left-Ctrl are down) or
1454     // AltGr-equivalent keys are down (any Ctrl key + any Alt key), and we have
1455     // a character (which indicates that there was an AltGr mapping), then act
1456     // as if alt and control are not really down for the purposes of modifiers.
1457     // This makes it so that if the user with, say, a German keyboard layout
1458     // presses AltGr-] (which we see as Right-Alt + Left-Ctrl + key), we just
1459     // output the key and we don't see the Alt and Ctrl keys.
1460     if (_is_ctrl_pressed(control_key_state) &&
1461         _is_alt_pressed(control_key_state)
1462         && (key_event->uChar.AsciiChar != '\0')) {
1463         // Try to remove as few bits as possible to improve our chances of
1464         // detecting combinations like Left-Alt + AltGr, Right-Ctrl + AltGr, or
1465         // Left-Alt + Right-Ctrl + AltGr.
1466         if ((control_key_state & RIGHT_ALT_PRESSED) != 0) {
1467             // Remove Right-Alt.
1468             control_key_state &= ~RIGHT_ALT_PRESSED;
1469             // If uChar is set, a Ctrl key is pressed, and Right-Alt is
1470             // pressed, Left-Ctrl is almost always set, except if the user
1471             // presses Right-Ctrl, then AltGr (in that specific order) for
1472             // whatever reason. At any rate, make sure the bit is not set.
1473             control_key_state &= ~LEFT_CTRL_PRESSED;
1474         } else if ((control_key_state & LEFT_ALT_PRESSED) != 0) {
1475             // Remove Left-Alt.
1476             control_key_state &= ~LEFT_ALT_PRESSED;
1477             // Whichever Ctrl key is down, remove it from the state. We only
1478             // remove one key, to improve our chances of detecting the
1479             // corner-case of Left-Ctrl + Left-Alt + Right-Ctrl.
1480             if ((control_key_state & LEFT_CTRL_PRESSED) != 0) {
1481                 // Remove Left-Ctrl.
1482                 control_key_state &= ~LEFT_CTRL_PRESSED;
1483             } else if ((control_key_state & RIGHT_CTRL_PRESSED) != 0) {
1484                 // Remove Right-Ctrl.
1485                 control_key_state &= ~RIGHT_CTRL_PRESSED;
1486             }
1487         }
1488 
1489         // Note that this logic isn't 100% perfect because Windows doesn't
1490         // allow us to detect all combinations because a physical AltGr key
1491         // press shows up as two bits, plus some combinations are ambiguous
1492         // about what is actually physically pressed.
1493     }
1494 
1495     return control_key_state;
1496 }
1497 
1498 // If NumLock is on and Shift is pressed, SHIFT_PRESSED is not set in
1499 // dwControlKeyState for the following keypad keys: period, 0-9. If we detect
1500 // this scenario, set the SHIFT_PRESSED bit so we can add modifiers
1501 // appropriately.
_normalize_keypad_control_key_state(const WORD vk,const DWORD control_key_state)1502 static DWORD _normalize_keypad_control_key_state(const WORD vk,
1503     const DWORD control_key_state) {
1504     if (!_is_numlock_on(control_key_state)) {
1505         return control_key_state;
1506     }
1507     if (!_is_enhanced_key(control_key_state)) {
1508         switch (vk) {
1509             case VK_INSERT: // 0
1510             case VK_DELETE: // .
1511             case VK_END:    // 1
1512             case VK_DOWN:   // 2
1513             case VK_NEXT:   // 3
1514             case VK_LEFT:   // 4
1515             case VK_CLEAR:  // 5
1516             case VK_RIGHT:  // 6
1517             case VK_HOME:   // 7
1518             case VK_UP:     // 8
1519             case VK_PRIOR:  // 9
1520                 return control_key_state | SHIFT_PRESSED;
1521         }
1522     }
1523 
1524     return control_key_state;
1525 }
1526 
_get_keypad_sequence(const DWORD control_key_state,const char * const normal,const char * const shifted)1527 static const char* _get_keypad_sequence(const DWORD control_key_state,
1528     const char* const normal, const char* const shifted) {
1529     if (_is_shift_pressed(control_key_state)) {
1530         // Shift is pressed and NumLock is off
1531         return shifted;
1532     } else {
1533         // Shift is not pressed and NumLock is off, or,
1534         // Shift is pressed and NumLock is on, in which case we want the
1535         // NumLock and Shift to neutralize each other, thus, we want the normal
1536         // sequence.
1537         return normal;
1538     }
1539     // If Shift is not pressed and NumLock is on, a different virtual key code
1540     // is returned by Windows, which can be taken care of by a different case
1541     // statement in _console_read().
1542 }
1543 
1544 // Write sequence to buf and return the number of bytes written.
_get_modifier_sequence(char * const buf,const WORD vk,DWORD control_key_state,const char * const normal)1545 static size_t _get_modifier_sequence(char* const buf, const WORD vk,
1546     DWORD control_key_state, const char* const normal) {
1547     // Copy the base sequence into buf.
1548     const size_t len = strlen(normal);
1549     memcpy(buf, normal, len);
1550 
1551     int code = 0;
1552 
1553     control_key_state = _normalize_keypad_control_key_state(vk,
1554         control_key_state);
1555 
1556     if (_is_shift_pressed(control_key_state)) {
1557         code |= 0x1;
1558     }
1559     if (_is_alt_pressed(control_key_state)) {   // any alt key pressed
1560         code |= 0x2;
1561     }
1562     if (_is_ctrl_pressed(control_key_state)) {  // any control key pressed
1563         code |= 0x4;
1564     }
1565     // If some modifier was held down, then we need to insert the modifier code
1566     if (code != 0) {
1567         if (len == 0) {
1568             // Should be impossible because caller should pass a string of
1569             // non-zero length.
1570             return 0;
1571         }
1572         size_t index = len - 1;
1573         const char lastChar = buf[index];
1574         if (lastChar != '~') {
1575             buf[index++] = '1';
1576         }
1577         buf[index++] = ';';         // modifier separator
1578         // 2 = shift, 3 = alt, 4 = shift & alt, 5 = control,
1579         // 6 = shift & control, 7 = alt & control, 8 = shift & alt & control
1580         buf[index++] = '1' + code;
1581         buf[index++] = lastChar;    // move ~ (or other last char) to the end
1582         return index;
1583     }
1584     return len;
1585 }
1586 
1587 // Write sequence to buf and return the number of bytes written.
_get_modifier_keypad_sequence(char * const buf,const WORD vk,const DWORD control_key_state,const char * const normal,const char shifted)1588 static size_t _get_modifier_keypad_sequence(char* const buf, const WORD vk,
1589     const DWORD control_key_state, const char* const normal,
1590     const char shifted) {
1591     if (_is_shift_pressed(control_key_state)) {
1592         // Shift is pressed and NumLock is off
1593         if (shifted != '\0') {
1594             buf[0] = shifted;
1595             return sizeof(buf[0]);
1596         } else {
1597             return 0;
1598         }
1599     } else {
1600         // Shift is not pressed and NumLock is off, or,
1601         // Shift is pressed and NumLock is on, in which case we want the
1602         // NumLock and Shift to neutralize each other, thus, we want the normal
1603         // sequence.
1604         return _get_modifier_sequence(buf, vk, control_key_state, normal);
1605     }
1606     // If Shift is not pressed and NumLock is on, a different virtual key code
1607     // is returned by Windows, which can be taken care of by a different case
1608     // statement in _console_read().
1609 }
1610 
1611 // The decimal key on the keypad produces a '.' for U.S. English and a ',' for
1612 // Standard German. Figure this out at runtime so we know what to output for
1613 // Shift-VK_DELETE.
_get_decimal_char()1614 static char _get_decimal_char() {
1615     return (char)MapVirtualKeyA(VK_DECIMAL, MAPVK_VK_TO_CHAR);
1616 }
1617 
1618 // Prefix the len bytes in buf with the escape character, and then return the
1619 // new buffer length.
_escape_prefix(char * const buf,const size_t len)1620 size_t _escape_prefix(char* const buf, const size_t len) {
1621     // If nothing to prefix, don't do anything. We might be called with
1622     // len == 0, if alt was held down with a dead key which produced nothing.
1623     if (len == 0) {
1624         return 0;
1625     }
1626 
1627     memmove(&buf[1], buf, len);
1628     buf[0] = '\x1b';
1629     return len + 1;
1630 }
1631 
1632 // Internal buffer to satisfy future _console_read() calls.
1633 static auto& g_console_input_buffer = *new std::vector<char>();
1634 
1635 // Writes to buffer buf (of length len), returning number of bytes written or -1 on error. Never
1636 // returns zero on console closure because Win32 consoles are never 'closed' (as far as I can tell).
_console_read(const HANDLE console,void * buf,size_t len)1637 static int _console_read(const HANDLE console, void* buf, size_t len) {
1638     for (;;) {
1639         // Read of zero bytes should not block waiting for something from the console.
1640         if (len == 0) {
1641             return 0;
1642         }
1643 
1644         // Flush as much as possible from input buffer.
1645         if (!g_console_input_buffer.empty()) {
1646             const int bytes_read = std::min(len, g_console_input_buffer.size());
1647             memcpy(buf, g_console_input_buffer.data(), bytes_read);
1648             const auto begin = g_console_input_buffer.begin();
1649             g_console_input_buffer.erase(begin, begin + bytes_read);
1650             return bytes_read;
1651         }
1652 
1653         // Read from the actual console. This may block until input.
1654         INPUT_RECORD input_record;
1655         if (!_get_key_event_record(console, &input_record)) {
1656             return -1;
1657         }
1658 
1659         KEY_EVENT_RECORD* const key_event = &input_record.Event.KeyEvent;
1660         const WORD vk = key_event->wVirtualKeyCode;
1661         const CHAR ch = key_event->uChar.AsciiChar;
1662         const DWORD control_key_state = _normalize_altgr_control_key_state(
1663             key_event);
1664 
1665         // The following emulation code should write the output sequence to
1666         // either seqstr or to seqbuf and seqbuflen.
1667         const char* seqstr = NULL;  // NULL terminated C-string
1668         // Enough space for max sequence string below, plus modifiers and/or
1669         // escape prefix.
1670         char seqbuf[16];
1671         size_t seqbuflen = 0;       // Space used in seqbuf.
1672 
1673 #define MATCH(vk, normal) \
1674             case (vk): \
1675             { \
1676                 seqstr = (normal); \
1677             } \
1678             break;
1679 
1680         // Modifier keys should affect the output sequence.
1681 #define MATCH_MODIFIER(vk, normal) \
1682             case (vk): \
1683             { \
1684                 seqbuflen = _get_modifier_sequence(seqbuf, (vk), \
1685                     control_key_state, (normal)); \
1686             } \
1687             break;
1688 
1689         // The shift key should affect the output sequence.
1690 #define MATCH_KEYPAD(vk, normal, shifted) \
1691             case (vk): \
1692             { \
1693                 seqstr = _get_keypad_sequence(control_key_state, (normal), \
1694                     (shifted)); \
1695             } \
1696             break;
1697 
1698         // The shift key and other modifier keys should affect the output
1699         // sequence.
1700 #define MATCH_MODIFIER_KEYPAD(vk, normal, shifted) \
1701             case (vk): \
1702             { \
1703                 seqbuflen = _get_modifier_keypad_sequence(seqbuf, (vk), \
1704                     control_key_state, (normal), (shifted)); \
1705             } \
1706             break;
1707 
1708 #define ESC "\x1b"
1709 #define CSI ESC "["
1710 #define SS3 ESC "O"
1711 
1712         // Only support normal mode, not application mode.
1713 
1714         // Enhanced keys:
1715         // * 6-pack: insert, delete, home, end, page up, page down
1716         // * cursor keys: up, down, right, left
1717         // * keypad: divide, enter
1718         // * Undocumented: VK_PAUSE (Ctrl-NumLock), VK_SNAPSHOT,
1719         //   VK_CANCEL (Ctrl-Pause/Break), VK_NUMLOCK
1720         if (_is_enhanced_key(control_key_state)) {
1721             switch (vk) {
1722                 case VK_RETURN: // Enter key on keypad
1723                     if (_is_ctrl_pressed(control_key_state)) {
1724                         seqstr = "\n";
1725                     } else {
1726                         seqstr = "\r";
1727                     }
1728                     break;
1729 
1730                 MATCH_MODIFIER(VK_PRIOR, CSI "5~"); // Page Up
1731                 MATCH_MODIFIER(VK_NEXT,  CSI "6~"); // Page Down
1732 
1733                 // gnome-terminal currently sends SS3 "F" and SS3 "H", but that
1734                 // will be fixed soon to match xterm which sends CSI "F" and
1735                 // CSI "H". https://bugzilla.redhat.com/show_bug.cgi?id=1119764
1736                 MATCH(VK_END,  CSI "F");
1737                 MATCH(VK_HOME, CSI "H");
1738 
1739                 MATCH_MODIFIER(VK_LEFT,  CSI "D");
1740                 MATCH_MODIFIER(VK_UP,    CSI "A");
1741                 MATCH_MODIFIER(VK_RIGHT, CSI "C");
1742                 MATCH_MODIFIER(VK_DOWN,  CSI "B");
1743 
1744                 MATCH_MODIFIER(VK_INSERT, CSI "2~");
1745                 MATCH_MODIFIER(VK_DELETE, CSI "3~");
1746 
1747                 MATCH(VK_DIVIDE, "/");
1748             }
1749         } else {    // Non-enhanced keys:
1750             switch (vk) {
1751                 case VK_BACK:   // backspace
1752                     if (_is_alt_pressed(control_key_state)) {
1753                         seqstr = ESC "\x7f";
1754                     } else {
1755                         seqstr = "\x7f";
1756                     }
1757                     break;
1758 
1759                 case VK_TAB:
1760                     if (_is_shift_pressed(control_key_state)) {
1761                         seqstr = CSI "Z";
1762                     } else {
1763                         seqstr = "\t";
1764                     }
1765                     break;
1766 
1767                 // Number 5 key in keypad when NumLock is off, or if NumLock is
1768                 // on and Shift is down.
1769                 MATCH_KEYPAD(VK_CLEAR, CSI "E", "5");
1770 
1771                 case VK_RETURN:     // Enter key on main keyboard
1772                     if (_is_alt_pressed(control_key_state)) {
1773                         seqstr = ESC "\n";
1774                     } else if (_is_ctrl_pressed(control_key_state)) {
1775                         seqstr = "\n";
1776                     } else {
1777                         seqstr = "\r";
1778                     }
1779                     break;
1780 
1781                 // VK_ESCAPE: Don't do any special handling. The OS uses many
1782                 // of the sequences with Escape and many of the remaining
1783                 // sequences don't produce bKeyDown messages, only !bKeyDown
1784                 // for whatever reason.
1785 
1786                 case VK_SPACE:
1787                     if (_is_alt_pressed(control_key_state)) {
1788                         seqstr = ESC " ";
1789                     } else if (_is_ctrl_pressed(control_key_state)) {
1790                         seqbuf[0] = '\0';   // NULL char
1791                         seqbuflen = 1;
1792                     } else {
1793                         seqstr = " ";
1794                     }
1795                     break;
1796 
1797                 MATCH_MODIFIER_KEYPAD(VK_PRIOR, CSI "5~", '9'); // Page Up
1798                 MATCH_MODIFIER_KEYPAD(VK_NEXT,  CSI "6~", '3'); // Page Down
1799 
1800                 MATCH_KEYPAD(VK_END,  CSI "4~", "1");
1801                 MATCH_KEYPAD(VK_HOME, CSI "1~", "7");
1802 
1803                 MATCH_MODIFIER_KEYPAD(VK_LEFT,  CSI "D", '4');
1804                 MATCH_MODIFIER_KEYPAD(VK_UP,    CSI "A", '8');
1805                 MATCH_MODIFIER_KEYPAD(VK_RIGHT, CSI "C", '6');
1806                 MATCH_MODIFIER_KEYPAD(VK_DOWN,  CSI "B", '2');
1807 
1808                 MATCH_MODIFIER_KEYPAD(VK_INSERT, CSI "2~", '0');
1809                 MATCH_MODIFIER_KEYPAD(VK_DELETE, CSI "3~",
1810                     _get_decimal_char());
1811 
1812                 case 0x30:          // 0
1813                 case 0x31:          // 1
1814                 case 0x39:          // 9
1815                 case VK_OEM_1:      // ;:
1816                 case VK_OEM_PLUS:   // =+
1817                 case VK_OEM_COMMA:  // ,<
1818                 case VK_OEM_PERIOD: // .>
1819                 case VK_OEM_7:      // '"
1820                 case VK_OEM_102:    // depends on keyboard, could be <> or \|
1821                 case VK_OEM_2:      // /?
1822                 case VK_OEM_3:      // `~
1823                 case VK_OEM_4:      // [{
1824                 case VK_OEM_5:      // \|
1825                 case VK_OEM_6:      // ]}
1826                 {
1827                     seqbuflen = _get_control_character(seqbuf, key_event,
1828                         control_key_state);
1829 
1830                     if (_is_alt_pressed(control_key_state)) {
1831                         seqbuflen = _escape_prefix(seqbuf, seqbuflen);
1832                     }
1833                 }
1834                 break;
1835 
1836                 case 0x32:          // 2
1837                 case 0x33:          // 3
1838                 case 0x34:          // 4
1839                 case 0x35:          // 5
1840                 case 0x36:          // 6
1841                 case 0x37:          // 7
1842                 case 0x38:          // 8
1843                 case VK_OEM_MINUS:  // -_
1844                 {
1845                     seqbuflen = _get_control_character(seqbuf, key_event,
1846                         control_key_state);
1847 
1848                     // If Alt is pressed and it isn't Ctrl-Alt-ShiftUp, then
1849                     // prefix with escape.
1850                     if (_is_alt_pressed(control_key_state) &&
1851                         !(_is_ctrl_pressed(control_key_state) &&
1852                         !_is_shift_pressed(control_key_state))) {
1853                         seqbuflen = _escape_prefix(seqbuf, seqbuflen);
1854                     }
1855                 }
1856                 break;
1857 
1858                 case 0x41:  // a
1859                 case 0x42:  // b
1860                 case 0x43:  // c
1861                 case 0x44:  // d
1862                 case 0x45:  // e
1863                 case 0x46:  // f
1864                 case 0x47:  // g
1865                 case 0x48:  // h
1866                 case 0x49:  // i
1867                 case 0x4a:  // j
1868                 case 0x4b:  // k
1869                 case 0x4c:  // l
1870                 case 0x4d:  // m
1871                 case 0x4e:  // n
1872                 case 0x4f:  // o
1873                 case 0x50:  // p
1874                 case 0x51:  // q
1875                 case 0x52:  // r
1876                 case 0x53:  // s
1877                 case 0x54:  // t
1878                 case 0x55:  // u
1879                 case 0x56:  // v
1880                 case 0x57:  // w
1881                 case 0x58:  // x
1882                 case 0x59:  // y
1883                 case 0x5a:  // z
1884                 {
1885                     seqbuflen = _get_non_alt_char(seqbuf, key_event,
1886                         control_key_state);
1887 
1888                     // If Alt is pressed, then prefix with escape.
1889                     if (_is_alt_pressed(control_key_state)) {
1890                         seqbuflen = _escape_prefix(seqbuf, seqbuflen);
1891                     }
1892                 }
1893                 break;
1894 
1895                 // These virtual key codes are generated by the keys on the
1896                 // keypad *when NumLock is on* and *Shift is up*.
1897                 MATCH(VK_NUMPAD0, "0");
1898                 MATCH(VK_NUMPAD1, "1");
1899                 MATCH(VK_NUMPAD2, "2");
1900                 MATCH(VK_NUMPAD3, "3");
1901                 MATCH(VK_NUMPAD4, "4");
1902                 MATCH(VK_NUMPAD5, "5");
1903                 MATCH(VK_NUMPAD6, "6");
1904                 MATCH(VK_NUMPAD7, "7");
1905                 MATCH(VK_NUMPAD8, "8");
1906                 MATCH(VK_NUMPAD9, "9");
1907 
1908                 MATCH(VK_MULTIPLY, "*");
1909                 MATCH(VK_ADD,      "+");
1910                 MATCH(VK_SUBTRACT, "-");
1911                 // VK_DECIMAL is generated by the . key on the keypad *when
1912                 // NumLock is on* and *Shift is up* and the sequence is not
1913                 // Ctrl-Alt-NoShift-. (which causes Ctrl-Alt-Del and the
1914                 // Windows Security screen to come up).
1915                 case VK_DECIMAL:
1916                     // U.S. English uses '.', Germany German uses ','.
1917                     seqbuflen = _get_non_control_char(seqbuf, key_event,
1918                         control_key_state);
1919                     break;
1920 
1921                 MATCH_MODIFIER(VK_F1,  SS3 "P");
1922                 MATCH_MODIFIER(VK_F2,  SS3 "Q");
1923                 MATCH_MODIFIER(VK_F3,  SS3 "R");
1924                 MATCH_MODIFIER(VK_F4,  SS3 "S");
1925                 MATCH_MODIFIER(VK_F5,  CSI "15~");
1926                 MATCH_MODIFIER(VK_F6,  CSI "17~");
1927                 MATCH_MODIFIER(VK_F7,  CSI "18~");
1928                 MATCH_MODIFIER(VK_F8,  CSI "19~");
1929                 MATCH_MODIFIER(VK_F9,  CSI "20~");
1930                 MATCH_MODIFIER(VK_F10, CSI "21~");
1931                 MATCH_MODIFIER(VK_F11, CSI "23~");
1932                 MATCH_MODIFIER(VK_F12, CSI "24~");
1933 
1934                 MATCH_MODIFIER(VK_F13, CSI "25~");
1935                 MATCH_MODIFIER(VK_F14, CSI "26~");
1936                 MATCH_MODIFIER(VK_F15, CSI "28~");
1937                 MATCH_MODIFIER(VK_F16, CSI "29~");
1938                 MATCH_MODIFIER(VK_F17, CSI "31~");
1939                 MATCH_MODIFIER(VK_F18, CSI "32~");
1940                 MATCH_MODIFIER(VK_F19, CSI "33~");
1941                 MATCH_MODIFIER(VK_F20, CSI "34~");
1942 
1943                 // MATCH_MODIFIER(VK_F21, ???);
1944                 // MATCH_MODIFIER(VK_F22, ???);
1945                 // MATCH_MODIFIER(VK_F23, ???);
1946                 // MATCH_MODIFIER(VK_F24, ???);
1947             }
1948         }
1949 
1950 #undef MATCH
1951 #undef MATCH_MODIFIER
1952 #undef MATCH_KEYPAD
1953 #undef MATCH_MODIFIER_KEYPAD
1954 #undef ESC
1955 #undef CSI
1956 #undef SS3
1957 
1958         const char* out;
1959         size_t outlen;
1960 
1961         // Check for output in any of:
1962         // * seqstr is set (and strlen can be used to determine the length).
1963         // * seqbuf and seqbuflen are set
1964         // Fallback to ch from Windows.
1965         if (seqstr != NULL) {
1966             out = seqstr;
1967             outlen = strlen(seqstr);
1968         } else if (seqbuflen > 0) {
1969             out = seqbuf;
1970             outlen = seqbuflen;
1971         } else if (ch != '\0') {
1972             // Use whatever Windows told us it is.
1973             seqbuf[0] = ch;
1974             seqbuflen = 1;
1975             out = seqbuf;
1976             outlen = seqbuflen;
1977         } else {
1978             // No special handling for the virtual key code and Windows isn't
1979             // telling us a character code, then we don't know how to translate
1980             // the key press.
1981             //
1982             // Consume the input and 'continue' to cause us to get a new key
1983             // event.
1984             D("_console_read: unknown virtual key code: %d, enhanced: %s",
1985                 vk, _is_enhanced_key(control_key_state) ? "true" : "false");
1986             continue;
1987         }
1988 
1989         // put output wRepeatCount times into g_console_input_buffer
1990         while (key_event->wRepeatCount-- > 0) {
1991             g_console_input_buffer.insert(g_console_input_buffer.end(), out, out + outlen);
1992         }
1993 
1994         // Loop around and try to flush g_console_input_buffer
1995     }
1996 }
1997 
1998 static DWORD _old_console_mode; // previous GetConsoleMode() result
1999 static HANDLE _console_handle;  // when set, console mode should be restored
2000 
stdin_raw_init()2001 void stdin_raw_init() {
2002     const HANDLE in = _get_console_handle(STDIN_FILENO, &_old_console_mode);
2003     if (in == nullptr) {
2004         return;
2005     }
2006 
2007     // Disable ENABLE_PROCESSED_INPUT so that Ctrl-C is read instead of
2008     // calling the process Ctrl-C routine (configured by
2009     // SetConsoleCtrlHandler()).
2010     // Disable ENABLE_LINE_INPUT so that input is immediately sent.
2011     // Disable ENABLE_ECHO_INPUT to disable local echo. Disabling this
2012     // flag also seems necessary to have proper line-ending processing.
2013     DWORD new_console_mode = _old_console_mode & ~(ENABLE_PROCESSED_INPUT |
2014                                                    ENABLE_LINE_INPUT |
2015                                                    ENABLE_ECHO_INPUT);
2016     // Enable ENABLE_WINDOW_INPUT to get window resizes.
2017     new_console_mode |= ENABLE_WINDOW_INPUT;
2018 
2019     if (!SetConsoleMode(in, new_console_mode)) {
2020         // This really should not fail.
2021         D("stdin_raw_init: SetConsoleMode() failed: %s",
2022           android::base::SystemErrorCodeToString(GetLastError()).c_str());
2023     }
2024 
2025     // Once this is set, it means that stdin has been configured for
2026     // reading from and that the old console mode should be restored later.
2027     _console_handle = in;
2028 
2029     // Note that we don't need to configure C Runtime line-ending
2030     // translation because _console_read() does not call the C Runtime to
2031     // read from the console.
2032 }
2033 
stdin_raw_restore()2034 void stdin_raw_restore() {
2035     if (_console_handle != NULL) {
2036         const HANDLE in = _console_handle;
2037         _console_handle = NULL;  // clear state
2038 
2039         if (!SetConsoleMode(in, _old_console_mode)) {
2040             // This really should not fail.
2041             D("stdin_raw_restore: SetConsoleMode() failed: %s",
2042               android::base::SystemErrorCodeToString(GetLastError()).c_str());
2043         }
2044     }
2045 }
2046 
2047 // Called by 'adb shell' and 'adb exec-in' (via unix_read()) to read from stdin.
unix_read_interruptible(int fd,void * buf,size_t len)2048 int unix_read_interruptible(int fd, void* buf, size_t len) {
2049     if ((fd == STDIN_FILENO) && (_console_handle != NULL)) {
2050         // If it is a request to read from stdin, and stdin_raw_init() has been
2051         // called, and it successfully configured the console, then read from
2052         // the console using Win32 console APIs and partially emulate a unix
2053         // terminal.
2054         return _console_read(_console_handle, buf, len);
2055     } else {
2056         // On older versions of Windows (definitely 7, definitely not 10),
2057         // ReadConsole() with a size >= 31367 fails, so if |fd| is a console
2058         // we need to limit the read size.
2059         if (len > 4096 && unix_isatty(fd)) {
2060             len = 4096;
2061         }
2062         // Just call into C Runtime which can read from pipes/files and which
2063         // can do LF/CR translation (which is overridable with _setmode()).
2064         // Undefine the macro that is set in sysdeps.h which bans calls to
2065         // plain read() in favor of unix_read() or adb_read().
2066 #pragma push_macro("read")
2067 #undef read
2068         return read(fd, buf, len);
2069 #pragma pop_macro("read")
2070     }
2071 }
2072 
2073 /**************************************************************************/
2074 /**************************************************************************/
2075 /*****                                                                *****/
2076 /*****      Unicode support                                           *****/
2077 /*****                                                                *****/
2078 /**************************************************************************/
2079 /**************************************************************************/
2080 
2081 // This implements support for using files with Unicode filenames and for
2082 // outputting Unicode text to a Win32 console window. This is inspired from
2083 // http://utf8everywhere.org/.
2084 //
2085 // Background
2086 // ----------
2087 //
2088 // On POSIX systems, to deal with files with Unicode filenames, just pass UTF-8
2089 // filenames to APIs such as open(). This works because filenames are largely
2090 // opaque 'cookies' (perhaps excluding path separators).
2091 //
2092 // On Windows, the native file APIs such as CreateFileW() take 2-byte wchar_t
2093 // UTF-16 strings. There is an API, CreateFileA() that takes 1-byte char
2094 // strings, but the strings are in the ANSI codepage and not UTF-8. (The
2095 // CreateFile() API is really just a macro that adds the W/A based on whether
2096 // the UNICODE preprocessor symbol is defined).
2097 //
2098 // Options
2099 // -------
2100 //
2101 // Thus, to write a portable program, there are a few options:
2102 //
2103 // 1. Write the program with wchar_t filenames (wchar_t path[256];).
2104 //    For Windows, just call CreateFileW(). For POSIX, write a wrapper openW()
2105 //    that takes a wchar_t string, converts it to UTF-8 and then calls the real
2106 //    open() API.
2107 //
2108 // 2. Write the program with a TCHAR typedef that is 2 bytes on Windows and
2109 //    1 byte on POSIX. Make T-* wrappers for various OS APIs and call those,
2110 //    potentially touching a lot of code.
2111 //
2112 // 3. Write the program with a 1-byte char filenames (char path[256];) that are
2113 //    UTF-8. For POSIX, just call open(). For Windows, write a wrapper that
2114 //    takes a UTF-8 string, converts it to UTF-16 and then calls the real OS
2115 //    or C Runtime API.
2116 //
2117 // The Choice
2118 // ----------
2119 //
2120 // The code below chooses option 3, the UTF-8 everywhere strategy. It uses
2121 // android::base::WideToUTF8() which converts UTF-16 to UTF-8. This is used by the
2122 // NarrowArgs helper class that is used to convert wmain() args into UTF-8
2123 // args that are passed to main() at the beginning of program startup. We also use
2124 // android::base::UTF8ToWide() which converts from UTF-8 to UTF-16. This is used to
2125 // implement wrappers below that call UTF-16 OS and C Runtime APIs.
2126 //
2127 // Unicode console output
2128 // ----------------------
2129 //
2130 // The way to output Unicode to a Win32 console window is to call
2131 // WriteConsoleW() with UTF-16 text. (The user must also choose a proper font
2132 // such as Lucida Console or Consolas, and in the case of East Asian languages
2133 // (such as Chinese, Japanese, Korean), the user must go to the Control Panel
2134 // and change the "system locale" to Chinese, etc., which allows a Chinese, etc.
2135 // font to be used in console windows.)
2136 //
2137 // The problem is getting the C Runtime to make fprintf and related APIs call
2138 // WriteConsoleW() under the covers. The C Runtime API, _setmode() sounds
2139 // promising, but the various modes have issues:
2140 //
2141 // 1. _setmode(_O_TEXT) (the default) does not use WriteConsoleW() so UTF-8 and
2142 //    UTF-16 do not display properly.
2143 // 2. _setmode(_O_BINARY) does not use WriteConsoleW() and the text comes out
2144 //    totally wrong.
2145 // 3. _setmode(_O_U8TEXT) seems to cause the C Runtime _invalid_parameter
2146 //    handler to be called (upon a later I/O call), aborting the process.
2147 // 4. _setmode(_O_U16TEXT) and _setmode(_O_WTEXT) cause non-wide printf/fprintf
2148 //    to output nothing.
2149 //
2150 // So the only solution is to write our own adb_fprintf() that converts UTF-8
2151 // to UTF-16 and then calls WriteConsoleW().
2152 
2153 
2154 // Constructor for helper class to convert wmain() UTF-16 args to UTF-8 to
2155 // be passed to main().
NarrowArgs(const int argc,wchar_t ** const argv)2156 NarrowArgs::NarrowArgs(const int argc, wchar_t** const argv) {
2157     narrow_args = new char*[argc + 1];
2158 
2159     for (int i = 0; i < argc; ++i) {
2160         std::string arg_narrow;
2161         if (!android::base::WideToUTF8(argv[i], &arg_narrow)) {
2162             fatal_errno("cannot convert argument from UTF-16 to UTF-8");
2163         }
2164         narrow_args[i] = strdup(arg_narrow.c_str());
2165     }
2166     narrow_args[argc] = nullptr;   // terminate
2167 }
2168 
~NarrowArgs()2169 NarrowArgs::~NarrowArgs() {
2170     if (narrow_args != nullptr) {
2171         for (char** argp = narrow_args; *argp != nullptr; ++argp) {
2172             free(*argp);
2173         }
2174         delete[] narrow_args;
2175         narrow_args = nullptr;
2176     }
2177 }
2178 
unix_open(const char * path,int options,...)2179 int unix_open(const char* path, int options, ...) {
2180     std::wstring path_wide;
2181     if (!android::base::UTF8ToWide(path, &path_wide)) {
2182         return -1;
2183     }
2184     if ((options & O_CREAT) == 0) {
2185         return _wopen(path_wide.c_str(), options);
2186     } else {
2187         int      mode;
2188         va_list  args;
2189         va_start(args, options);
2190         mode = va_arg(args, int);
2191         va_end(args);
2192         return _wopen(path_wide.c_str(), options, mode);
2193     }
2194 }
2195 
2196 // Version of opendir() that takes a UTF-8 path.
adb_opendir(const char * path)2197 DIR* adb_opendir(const char* path) {
2198     std::wstring path_wide;
2199     if (!android::base::UTF8ToWide(path, &path_wide)) {
2200         return nullptr;
2201     }
2202 
2203     // Just cast _WDIR* to DIR*. This doesn't work if the caller reads any of
2204     // the fields, but right now all the callers treat the structure as
2205     // opaque.
2206     return reinterpret_cast<DIR*>(_wopendir(path_wide.c_str()));
2207 }
2208 
2209 // Version of readdir() that returns UTF-8 paths.
adb_readdir(DIR * dir)2210 struct dirent* adb_readdir(DIR* dir) {
2211     _WDIR* const wdir = reinterpret_cast<_WDIR*>(dir);
2212     struct _wdirent* const went = _wreaddir(wdir);
2213     if (went == nullptr) {
2214         return nullptr;
2215     }
2216 
2217     // Convert from UTF-16 to UTF-8.
2218     std::string name_utf8;
2219     if (!android::base::WideToUTF8(went->d_name, &name_utf8)) {
2220         return nullptr;
2221     }
2222 
2223     // Cast the _wdirent* to dirent* and overwrite the d_name field (which has
2224     // space for UTF-16 wchar_t's) with UTF-8 char's.
2225     struct dirent* ent = reinterpret_cast<struct dirent*>(went);
2226 
2227     if (name_utf8.length() + 1 > sizeof(went->d_name)) {
2228         // Name too big to fit in existing buffer.
2229         errno = ENOMEM;
2230         return nullptr;
2231     }
2232 
2233     // Note that sizeof(_wdirent::d_name) is bigger than sizeof(dirent::d_name)
2234     // because _wdirent contains wchar_t instead of char. So even if name_utf8
2235     // can fit in _wdirent::d_name, the resulting dirent::d_name field may be
2236     // bigger than the caller expects because they expect a dirent structure
2237     // which has a smaller d_name field. Ignore this since the caller should be
2238     // resilient.
2239 
2240     // Rewrite the UTF-16 d_name field to UTF-8.
2241     strcpy(ent->d_name, name_utf8.c_str());
2242 
2243     return ent;
2244 }
2245 
2246 // Version of closedir() to go with our version of adb_opendir().
adb_closedir(DIR * dir)2247 int adb_closedir(DIR* dir) {
2248     return _wclosedir(reinterpret_cast<_WDIR*>(dir));
2249 }
2250 
2251 // Version of unlink() that takes a UTF-8 path.
adb_unlink(const char * path)2252 int adb_unlink(const char* path) {
2253     std::wstring wpath;
2254     if (!android::base::UTF8ToWide(path, &wpath)) {
2255         return -1;
2256     }
2257 
2258     int  rc = _wunlink(wpath.c_str());
2259 
2260     if (rc == -1 && errno == EACCES) {
2261         /* unlink returns EACCES when the file is read-only, so we first */
2262         /* try to make it writable, then unlink again...                 */
2263         rc = _wchmod(wpath.c_str(), _S_IREAD | _S_IWRITE);
2264         if (rc == 0)
2265             rc = _wunlink(wpath.c_str());
2266     }
2267     return rc;
2268 }
2269 
2270 // Version of mkdir() that takes a UTF-8 path.
adb_mkdir(const std::string & path,int mode)2271 int adb_mkdir(const std::string& path, int mode) {
2272     std::wstring path_wide;
2273     if (!android::base::UTF8ToWide(path, &path_wide)) {
2274         return -1;
2275     }
2276 
2277     return _wmkdir(path_wide.c_str());
2278 }
2279 
2280 // Version of utime() that takes a UTF-8 path.
adb_utime(const char * path,struct utimbuf * u)2281 int adb_utime(const char* path, struct utimbuf* u) {
2282     std::wstring path_wide;
2283     if (!android::base::UTF8ToWide(path, &path_wide)) {
2284         return -1;
2285     }
2286 
2287     static_assert(sizeof(struct utimbuf) == sizeof(struct _utimbuf),
2288         "utimbuf and _utimbuf should be the same size because they both "
2289         "contain the same types, namely time_t");
2290     return _wutime(path_wide.c_str(), reinterpret_cast<struct _utimbuf*>(u));
2291 }
2292 
2293 // Version of chmod() that takes a UTF-8 path.
adb_chmod(const char * path,int mode)2294 int adb_chmod(const char* path, int mode) {
2295     std::wstring path_wide;
2296     if (!android::base::UTF8ToWide(path, &path_wide)) {
2297         return -1;
2298     }
2299 
2300     return _wchmod(path_wide.c_str(), mode);
2301 }
2302 
2303 // From libutils/Unicode.cpp, get the length of a UTF-8 sequence given the lead byte.
utf8_codepoint_len(uint8_t ch)2304 static inline size_t utf8_codepoint_len(uint8_t ch) {
2305     return ((0xe5000000 >> ((ch >> 3) & 0x1e)) & 3) + 1;
2306 }
2307 
2308 namespace internal {
2309 
2310 // Given a sequence of UTF-8 bytes (denoted by the range [first, last)), return the number of bytes
2311 // (from the beginning) that are complete UTF-8 sequences and append the remaining bytes to
2312 // remaining_bytes.
ParseCompleteUTF8(const char * const first,const char * const last,std::vector<char> * const remaining_bytes)2313 size_t ParseCompleteUTF8(const char* const first, const char* const last,
2314                          std::vector<char>* const remaining_bytes) {
2315     // Walk backwards from the end of the sequence looking for the beginning of a UTF-8 sequence.
2316     // Current_after points one byte past the current byte to be examined.
2317     for (const char* current_after = last; current_after != first; --current_after) {
2318         const char* const current = current_after - 1;
2319         const char ch = *current;
2320         const char kHighBit = 0x80u;
2321         const char kTwoHighestBits = 0xC0u;
2322         if ((ch & kHighBit) == 0) { // high bit not set
2323             // The buffer ends with a one-byte UTF-8 sequence, possibly followed by invalid trailing
2324             // bytes with no leading byte, so return the entire buffer.
2325             break;
2326         } else if ((ch & kTwoHighestBits) == kTwoHighestBits) { // top two highest bits set
2327             // Lead byte in UTF-8 sequence, so check if we have all the bytes in the sequence.
2328             const size_t bytes_available = last - current;
2329             if (bytes_available < utf8_codepoint_len(ch)) {
2330                 // We don't have all the bytes in the UTF-8 sequence, so return all the bytes
2331                 // preceding the current incomplete UTF-8 sequence and append the remaining bytes
2332                 // to remaining_bytes.
2333                 remaining_bytes->insert(remaining_bytes->end(), current, last);
2334                 return current - first;
2335             } else {
2336                 // The buffer ends with a complete UTF-8 sequence, possibly followed by invalid
2337                 // trailing bytes with no lead byte, so return the entire buffer.
2338                 break;
2339             }
2340         } else {
2341             // Trailing byte, so keep going backwards looking for the lead byte.
2342         }
2343     }
2344 
2345     // Return the size of the entire buffer. It is possible that we walked backward past invalid
2346     // trailing bytes with no lead byte, in which case we want to return all those invalid bytes
2347     // so that they can be processed.
2348     return last - first;
2349 }
2350 
2351 }
2352 
2353 // Bytes that have not yet been output to the console because they are incomplete UTF-8 sequences.
2354 // Note that we use only one buffer even though stderr and stdout are logically separate streams.
2355 // This matches the behavior of Linux.
2356 
2357 // Internal helper function to write UTF-8 bytes to a console. Returns -1 on error.
_console_write_utf8(const char * const buf,const size_t buf_size,FILE * stream,HANDLE console)2358 static int _console_write_utf8(const char* const buf, const size_t buf_size, FILE* stream,
2359                                HANDLE console) {
2360     static std::mutex& console_output_buffer_lock = *new std::mutex();
2361     static auto& console_output_buffer = *new std::vector<char>();
2362 
2363     const int saved_errno = errno;
2364     std::vector<char> combined_buffer;
2365 
2366     // Complete UTF-8 sequences that should be immediately written to the console.
2367     const char* utf8;
2368     size_t utf8_size;
2369 
2370     {
2371         std::lock_guard<std::mutex> lock(console_output_buffer_lock);
2372         if (console_output_buffer.empty()) {
2373             // If console_output_buffer doesn't have a buffered up incomplete UTF-8 sequence (the
2374             // common case with plain ASCII), parse buf directly.
2375             utf8 = buf;
2376             utf8_size = internal::ParseCompleteUTF8(buf, buf + buf_size, &console_output_buffer);
2377         } else {
2378             // If console_output_buffer has a buffered up incomplete UTF-8 sequence, move it to
2379             // combined_buffer (and effectively clear console_output_buffer) and append buf to
2380             // combined_buffer, then parse it all together.
2381             combined_buffer.swap(console_output_buffer);
2382             combined_buffer.insert(combined_buffer.end(), buf, buf + buf_size);
2383 
2384             utf8 = combined_buffer.data();
2385             utf8_size = internal::ParseCompleteUTF8(utf8, utf8 + combined_buffer.size(),
2386                                                     &console_output_buffer);
2387         }
2388     }
2389 
2390     std::wstring utf16;
2391 
2392     // Try to convert from data that might be UTF-8 to UTF-16, ignoring errors (just like Linux
2393     // which does not return an error on bad UTF-8). Data might not be UTF-8 if the user cat's
2394     // random data, runs dmesg (which might have non-UTF-8), etc.
2395     // This could throw std::bad_alloc.
2396     (void)android::base::UTF8ToWide(utf8, utf8_size, &utf16);
2397 
2398     // Note that this does not do \n => \r\n translation because that
2399     // doesn't seem necessary for the Windows console. For the Windows
2400     // console \r moves to the beginning of the line and \n moves to a new
2401     // line.
2402 
2403     // Flush any stream buffering so that our output is afterwards which
2404     // makes sense because our call is afterwards.
2405     (void)fflush(stream);
2406 
2407     // Write UTF-16 to the console.
2408     DWORD written = 0;
2409     if (!WriteConsoleW(console, utf16.c_str(), utf16.length(), &written, NULL)) {
2410         errno = EIO;
2411         return -1;
2412     }
2413 
2414     // Return the size of the original buffer passed in, signifying that we consumed it all, even
2415     // if nothing was displayed, in the case of being passed an incomplete UTF-8 sequence. This
2416     // matches the Linux behavior.
2417     errno = saved_errno;
2418     return buf_size;
2419 }
2420 
2421 // Function prototype because attributes cannot be placed on func definitions.
2422 static int _console_vfprintf(const HANDLE console, FILE* stream,
2423                              const char *format, va_list ap)
2424     __attribute__((__format__(ADB_FORMAT_ARCHETYPE, 3, 0)));
2425 
2426 // Internal function to format a UTF-8 string and write it to a Win32 console.
2427 // Returns -1 on error.
_console_vfprintf(const HANDLE console,FILE * stream,const char * format,va_list ap)2428 static int _console_vfprintf(const HANDLE console, FILE* stream,
2429                              const char *format, va_list ap) {
2430     const int saved_errno = errno;
2431     std::string output_utf8;
2432 
2433     // Format the string.
2434     // This could throw std::bad_alloc.
2435     android::base::StringAppendV(&output_utf8, format, ap);
2436 
2437     const int result = _console_write_utf8(output_utf8.c_str(), output_utf8.length(), stream,
2438                                            console);
2439     if (result != -1) {
2440         errno = saved_errno;
2441     } else {
2442         // If -1 was returned, errno has been set.
2443     }
2444     return result;
2445 }
2446 
2447 // Version of vfprintf() that takes UTF-8 and can write Unicode to a
2448 // Windows console.
adb_vfprintf(FILE * stream,const char * format,va_list ap)2449 int adb_vfprintf(FILE *stream, const char *format, va_list ap) {
2450     const HANDLE console = _get_console_handle(stream);
2451 
2452     // If there is an associated Win32 console, write to it specially,
2453     // otherwise defer to the regular C Runtime, passing it UTF-8.
2454     if (console != NULL) {
2455         return _console_vfprintf(console, stream, format, ap);
2456     } else {
2457         // If vfprintf is a macro, undefine it, so we can call the real
2458         // C Runtime API.
2459 #pragma push_macro("vfprintf")
2460 #undef vfprintf
2461         return vfprintf(stream, format, ap);
2462 #pragma pop_macro("vfprintf")
2463     }
2464 }
2465 
2466 // Version of vprintf() that takes UTF-8 and can write Unicode to a Windows console.
adb_vprintf(const char * format,va_list ap)2467 int adb_vprintf(const char *format, va_list ap) {
2468     return adb_vfprintf(stdout, format, ap);
2469 }
2470 
2471 // Version of fprintf() that takes UTF-8 and can write Unicode to a
2472 // Windows console.
adb_fprintf(FILE * stream,const char * format,...)2473 int adb_fprintf(FILE *stream, const char *format, ...) {
2474     va_list ap;
2475     va_start(ap, format);
2476     const int result = adb_vfprintf(stream, format, ap);
2477     va_end(ap);
2478 
2479     return result;
2480 }
2481 
2482 // Version of printf() that takes UTF-8 and can write Unicode to a
2483 // Windows console.
adb_printf(const char * format,...)2484 int adb_printf(const char *format, ...) {
2485     va_list ap;
2486     va_start(ap, format);
2487     const int result = adb_vfprintf(stdout, format, ap);
2488     va_end(ap);
2489 
2490     return result;
2491 }
2492 
2493 // Version of fputs() that takes UTF-8 and can write Unicode to a
2494 // Windows console.
adb_fputs(const char * buf,FILE * stream)2495 int adb_fputs(const char* buf, FILE* stream) {
2496     // adb_fprintf returns -1 on error, which is conveniently the same as EOF
2497     // which fputs (and hence adb_fputs) should return on error.
2498     static_assert(EOF == -1, "EOF is not -1, so this code needs to be fixed");
2499     return adb_fprintf(stream, "%s", buf);
2500 }
2501 
2502 // Version of fputc() that takes UTF-8 and can write Unicode to a
2503 // Windows console.
adb_fputc(int ch,FILE * stream)2504 int adb_fputc(int ch, FILE* stream) {
2505     const int result = adb_fprintf(stream, "%c", ch);
2506     if (result == -1) {
2507         return EOF;
2508     }
2509     // For success, fputc returns the char, cast to unsigned char, then to int.
2510     return static_cast<unsigned char>(ch);
2511 }
2512 
2513 // Version of putchar() that takes UTF-8 and can write Unicode to a Windows console.
adb_putchar(int ch)2514 int adb_putchar(int ch) {
2515     return adb_fputc(ch, stdout);
2516 }
2517 
2518 // Version of puts() that takes UTF-8 and can write Unicode to a Windows console.
adb_puts(const char * buf)2519 int adb_puts(const char* buf) {
2520     // adb_printf returns -1 on error, which is conveniently the same as EOF
2521     // which puts (and hence adb_puts) should return on error.
2522     static_assert(EOF == -1, "EOF is not -1, so this code needs to be fixed");
2523     return adb_printf("%s\n", buf);
2524 }
2525 
2526 // Internal function to write UTF-8 to a Win32 console. Returns the number of
2527 // items (of length size) written. On error, returns a short item count or 0.
_console_fwrite(const void * ptr,size_t size,size_t nmemb,FILE * stream,HANDLE console)2528 static size_t _console_fwrite(const void* ptr, size_t size, size_t nmemb,
2529                               FILE* stream, HANDLE console) {
2530     const int result = _console_write_utf8(reinterpret_cast<const char*>(ptr), size * nmemb, stream,
2531                                            console);
2532     if (result == -1) {
2533         return 0;
2534     }
2535     return result / size;
2536 }
2537 
2538 // Version of fwrite() that takes UTF-8 and can write Unicode to a
2539 // Windows console.
adb_fwrite(const void * ptr,size_t size,size_t nmemb,FILE * stream)2540 size_t adb_fwrite(const void* ptr, size_t size, size_t nmemb, FILE* stream) {
2541     const HANDLE console = _get_console_handle(stream);
2542 
2543     // If there is an associated Win32 console, write to it specially,
2544     // otherwise defer to the regular C Runtime, passing it UTF-8.
2545     if (console != NULL) {
2546         return _console_fwrite(ptr, size, nmemb, stream, console);
2547     } else {
2548         // If fwrite is a macro, undefine it, so we can call the real
2549         // C Runtime API.
2550 #pragma push_macro("fwrite")
2551 #undef fwrite
2552         return fwrite(ptr, size, nmemb, stream);
2553 #pragma pop_macro("fwrite")
2554     }
2555 }
2556 
2557 // Version of fopen() that takes a UTF-8 filename and can access a file with
2558 // a Unicode filename.
adb_fopen(const char * path,const char * mode)2559 FILE* adb_fopen(const char* path, const char* mode) {
2560     std::wstring path_wide;
2561     if (!android::base::UTF8ToWide(path, &path_wide)) {
2562         return nullptr;
2563     }
2564 
2565     std::wstring mode_wide;
2566     if (!android::base::UTF8ToWide(mode, &mode_wide)) {
2567         return nullptr;
2568     }
2569 
2570     return _wfopen(path_wide.c_str(), mode_wide.c_str());
2571 }
2572 
2573 // Return a lowercase version of the argument. Uses C Runtime tolower() on
2574 // each byte which is not UTF-8 aware, and theoretically uses the current C
2575 // Runtime locale (which in practice is not changed, so this becomes a ASCII
2576 // conversion).
ToLower(const std::string & anycase)2577 static std::string ToLower(const std::string& anycase) {
2578     // copy string
2579     std::string str(anycase);
2580     // transform the copy
2581     std::transform(str.begin(), str.end(), str.begin(), tolower);
2582     return str;
2583 }
2584 
2585 extern "C" int main(int argc, char** argv);
2586 
2587 // Link with -municode to cause this wmain() to be used as the program
2588 // entrypoint. It will convert the args from UTF-16 to UTF-8 and call the
2589 // regular main() with UTF-8 args.
wmain(int argc,wchar_t ** argv)2590 extern "C" int wmain(int argc, wchar_t **argv) {
2591     // Convert args from UTF-16 to UTF-8 and pass that to main().
2592     NarrowArgs narrow_args(argc, argv);
2593     return main(argc, narrow_args.data());
2594 }
2595 
2596 // Shadow UTF-8 environment variable name/value pairs that are created from
2597 // _wenviron the first time that adb_getenv() is called. Note that this is not
2598 // currently updated if putenv, setenv, unsetenv are called. Note that no
2599 // thread synchronization is done, but we're called early enough in
2600 // single-threaded startup that things work ok.
2601 static auto& g_environ_utf8 = *new std::unordered_map<std::string, char*>();
2602 
2603 // Make sure that shadow UTF-8 environment variables are setup.
_ensure_env_setup()2604 static void _ensure_env_setup() {
2605     // If some name/value pairs exist, then we've already done the setup below.
2606     if (g_environ_utf8.size() != 0) {
2607         return;
2608     }
2609 
2610     if (_wenviron == nullptr) {
2611         // If _wenviron is null, then -municode probably wasn't used. That
2612         // linker flag will cause the entry point to setup _wenviron. It will
2613         // also require an implementation of wmain() (which we provide above).
2614         fatal("_wenviron is not set, did you link with -municode?");
2615     }
2616 
2617     // Read name/value pairs from UTF-16 _wenviron and write new name/value
2618     // pairs to UTF-8 g_environ_utf8. Note that it probably does not make sense
2619     // to use the D() macro here because that tracing only works if the
2620     // ADB_TRACE environment variable is setup, but that env var can't be read
2621     // until this code completes.
2622     for (wchar_t** env = _wenviron; *env != nullptr; ++env) {
2623         wchar_t* const equal = wcschr(*env, L'=');
2624         if (equal == nullptr) {
2625             // Malformed environment variable with no equal sign. Shouldn't
2626             // really happen, but we should be resilient to this.
2627             continue;
2628         }
2629 
2630         // If we encounter an error converting UTF-16, don't error-out on account of a single env
2631         // var because the program might never even read this particular variable.
2632         std::string name_utf8;
2633         if (!android::base::WideToUTF8(*env, equal - *env, &name_utf8)) {
2634             continue;
2635         }
2636 
2637         // Store lowercase name so that we can do case-insensitive searches.
2638         name_utf8 = ToLower(name_utf8);
2639 
2640         std::string value_utf8;
2641         if (!android::base::WideToUTF8(equal + 1, &value_utf8)) {
2642             continue;
2643         }
2644 
2645         char* const value_dup = strdup(value_utf8.c_str());
2646 
2647         // Don't overwrite a previus env var with the same name. In reality,
2648         // the system probably won't let two env vars with the same name exist
2649         // in _wenviron.
2650         g_environ_utf8.insert({name_utf8, value_dup});
2651     }
2652 }
2653 
2654 // Version of getenv() that takes a UTF-8 environment variable name and
2655 // retrieves a UTF-8 value. Case-insensitive to match getenv() on Windows.
adb_getenv(const char * name)2656 char* adb_getenv(const char* name) {
2657     _ensure_env_setup();
2658 
2659     // Case-insensitive search by searching for lowercase name in a map of
2660     // lowercase names.
2661     const auto it = g_environ_utf8.find(ToLower(std::string(name)));
2662     if (it == g_environ_utf8.end()) {
2663         return nullptr;
2664     }
2665 
2666     return it->second;
2667 }
2668 
2669 // Version of getcwd() that returns the current working directory in UTF-8.
adb_getcwd(char * buf,int size)2670 char* adb_getcwd(char* buf, int size) {
2671     wchar_t* wbuf = _wgetcwd(nullptr, 0);
2672     if (wbuf == nullptr) {
2673         return nullptr;
2674     }
2675 
2676     std::string buf_utf8;
2677     const bool narrow_result = android::base::WideToUTF8(wbuf, &buf_utf8);
2678     free(wbuf);
2679     wbuf = nullptr;
2680 
2681     if (!narrow_result) {
2682         return nullptr;
2683     }
2684 
2685     // If size was specified, make sure all the chars will fit.
2686     if (size != 0) {
2687         if (size < static_cast<int>(buf_utf8.length() + 1)) {
2688             errno = ERANGE;
2689             return nullptr;
2690         }
2691     }
2692 
2693     // If buf was not specified, allocate storage.
2694     if (buf == nullptr) {
2695         if (size == 0) {
2696             size = buf_utf8.length() + 1;
2697         }
2698         buf = reinterpret_cast<char*>(malloc(size));
2699         if (buf == nullptr) {
2700             return nullptr;
2701         }
2702     }
2703 
2704     // Destination buffer was allocated with enough space, or we've already
2705     // checked an existing buffer size for enough space.
2706     strcpy(buf, buf_utf8.c_str());
2707 
2708     return buf;
2709 }
2710